540010 A7 B7 五、發明說明（ί ) [技術領域] 本發明係關於計算圖像中之物體數量之視覺裝置及其 方法，詳言之，係關於計算攝像機或數位相機所拍攝之三 原色波長、可視光波長、紅外線波長、紫外線波長，及其 他所有電磁波中，任意頻帶所構成的圖像中之移動物體或 靜止物體數量。 [習知技術] 一直以來，開發有使用攝像機或數位相機將複數之物 體拍攝於一個畫面，使用數位技術計算畫面中物體數量的 裝置。作爲其代表例而言，有透過顯微鏡計算所拍攝之細 胞數量的裝置。由於將細胞之染色體染成紫色，因此藉切 出圖像中一定以上大小的紫色區域作爲一塊，即能計算細 胞數量。然而，當細胞之染色有困難或無法染色時，即不 易計算細胞數量。此係因一般來說細胞爲透明之故，是以 爲了將整個細胞作成一塊，顏色資訊並不太有效。當然， 若放大細胞圖像的話，雖能捕捉細胞核及線粒體 (mitochondria)之陰影，但這種情形是例外。因此，多數的 情形是利用光之折射、反射所投影出的細胞輪廓來生成邊 緣資訊。只要使用這種邊緣資訊，理論上，雖能將整個細 胞作成一塊，但實際上，由於邊緣資訊幾乎皆爲不完全者 ，因此須使用細胞之形狀與大小資訊來輔助邊緣資訊。而 且，爲了自該邊緣資訊將整個細胞作成一塊，必須施以塗 黑等計算量較多的圖像處理，當然邊緣資訊中不可有裂縫 。再者，若自細胞中僅選出移動的細胞加以切出的話，勢 3 (請先閱讀背面之注意事項再填寫本頁) 裝--------訂----- 辱 經濟部智慧財產局員工消費合作社印製 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 540010 A7 B7 五、發明說明（、） 必要進行光擧流量等之計算，結果，爲提高計算數之精度 必須要有昂貴的裝置，另一方面，若使用便宜的裝置的話 ，計算時間將會變的非常龐大。 然而，除了細胞等部分例外之外，一般的物體皆有顏 色。例如，大致說來，蝌蚪背面是深褐色，蘋果是紅色、 黃色、綠色，馬是黑色、灰色，烏鴉是黑色。因此，爲計 算該物體之數量，只要自圖像中挑出物體固有的顏色資訊 即可，但事實上並非這麼簡單。其一，係由於顏色資訊會 受到陽光或煦明的亮度，以及攝像機性能極大的影響。此 外，若在攝影環鏡中有類似顏色的物體的話，即不易區別 對象物體及其以外之物體。因此，一般係使用形狀或大小 等資訊自背景切割出對象物體後，再計算該物體的數量， 爲減少計算量，所使用的顏色資訊僅限於針對對象物體的 程度。假如，視覺裝置能捕捉顏色資訊的變化作爲物體之 動態，照明及攝像機的性能雖不至成爲太大的問題，但不 僅不易從物體的動態正確再現物體的形狀，又會產生須塗 黑被邊緣資訊所圍住之區域以確定物體領域的問題，活用 顏色資訊變化的視覺裝置之相關硏究，仍舊不足。 考慮上述事情，若能使視覺裝置根據顏色資訊的變化 等來捕捉物體的動態以生成邊緣資訊，自該邊緣資訊分離 出物體以作成一塊的話，即能不依賴物體之特徵與攝影環 境而計算物體的數量。此外，若能使物體本身振動、或使 攝像機振動、再不然使所拍攝的圖像振動，據以根據顏色 資訊生成邊緣資訊的話，即可期待視覺裝置即使在物體靜 4 本紙張尺度適用中國國家標準（CNS)A4規格（210 x 297公釐） (請先閱讀背面之注意事項再填寫本頁) •裝 訂---- 經濟部智慧財產局員工消費合作社印製 540010 -：______B7 五、發明說明（> ) 濟 部 智 慧 財 產 局 員 4 合 作 社 印 製 止的情況下，亦能計算物體數量。 此處’假設有能計算攝像機所拍攝之物體數量的前述 視覺裝置。當攝像機所拍攝之物體係靜止的話，視覺裝置 就能經常計算靜止物體的數量。但是，當物體在移動時， 視覺裝置僅能在攝像機拍攝移動物體期間，計算移動物體 的數量。若像培養皿中之細胞般，能預先特定靜止物體及 移動物體之位置的話當然沒有問題，但若係計算室內或室 外空間走動的人或動物時，或由於無法將整個屋內或屋外 空間全部拍入攝像機之畫角中，或因爲與攝像機之距離使 得人或動物在圖像中變大或變小，因此若固定攝相機的話 ’視覺裝置的用途就會受到限制。再者，由於視覺裝置必 須將室內裝飾與人或動物加以區別，因此爲了辨識物體需 要龐大的計算量。 考慮到上述事情，所期待的視覺裝置是這樣的。若移 動式攝像機能自室內或室外空間中搜索出人或動物等特定 之物體並僅拍攝該等物體，且視覺裝置能調整攝像機之倍 率’以使該等物體在圖像中成適當之大小，視覺裝置即能 容易辨識該等物體，亦能計算人或動物等無法預先特定其 位置之移動物體的數量。當然，當人或動物等躺著幾乎不 動時’視覺裝置亦能將人或動物與其他靜止物體加以區別 ，以計算人與動物等的數量。 因此’本發明之目的在於，以選擇性地選擇動態圖像 中之移動物體或所有物體之任一者所生成的邊緣資訊爲基 準’高速地計算移動物體或所有物體的數量。此外，其另 5 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 請 先 閱 讀 背 意 事540010 A7 B7 V. Description of the Invention (Technical Field) The present invention relates to a visual device and method for calculating the number of objects in an image. Specifically, it relates to the calculation of the three primary color wavelengths, The number of moving or stationary objects in an image composed of an arbitrary frequency band among light wavelengths, infrared wavelengths, ultraviolet wavelengths, and all other electromagnetic waves. [Known Technology] Conventionally, devices have been developed that use a video camera or a digital camera to capture multiple objects on one screen and use digital technology to count the number of objects in the screen. As a representative example, there is a device for counting the number of cells photographed through a microscope. Because the chromosomes of the cells are dyed purple, the number of cells can be calculated by cutting out a purple area of a certain size or more in the image. However, when it is difficult or impossible to stain the cells, it is not easy to calculate the number of cells. This is because the cells are generally transparent, in order to make the whole cell into one piece, the color information is not very effective. Of course, if the cell image is enlarged, the shadow of the nucleus and mitochondria can be captured, but this case is the exception. Therefore, in most cases, the edge information is generated using the cell outlines projected by the refraction and reflection of light. As long as this edge information is used, in theory, although the entire cell can be made into one piece, in fact, since the edge information is almost incomplete, the shape and size information of the cells must be used to assist the edge information. In addition, in order to make the whole cell into a piece from the edge information, it is necessary to perform image processing such as blackening, and of course, there must be no cracks in the edge information. Furthermore, if only the moving cells are selected from the cells and cut out, the potential 3 (Please read the precautions on the back before filling out this page). -------- Order ----- Ministry of Economy Printed by the Intellectual Property Bureau Staff Consumer Cooperatives The paper size is applicable to the Chinese National Standard (CNS) A4 (210 X 297 mm) 540010 A7 B7 V. Description of the invention (,) The calculation of light flux, etc. is necessary. As a result, The accuracy of the calculations requires an expensive device. On the other hand, if an inexpensive device is used, the calculation time will become very large. However, with the exception of parts such as cells, ordinary objects have color. For example, roughly speaking, the back of the badger is dark brown, apples are red, yellow, and green, horses are black, gray, and crows are black. Therefore, in order to calculate the number of objects, it is only necessary to pick out the inherent color information of the object from the image, but in fact it is not so simple. One is that the color information is greatly affected by sunlight or bright brightness, and the camera performance. In addition, if there are similarly colored objects in the photographic ring mirror, it is not easy to distinguish the target object and other objects. Therefore, the shape or size is usually used to cut out the object from the background and then calculate the number of the object. To reduce the amount of calculation, the color information used is limited to the degree of the object. If the visual device can capture the change of color information as the dynamics of the object, although the performance of the lighting and the camera will not be too big a problem, it will not only be difficult to correctly reproduce the shape of the object from the dynamics of the object, but it will also produce blackened edges. The area surrounded by information is used to identify problems in the field of objects, and the related research using visual devices that change color information is still insufficient. Considering the above, if a visual device can be made to capture the dynamics of an object to generate edge information based on changes in color information, etc., and separate the object from the edge information to make a piece, it can calculate the object independently of the characteristics of the object and the shooting environment. quantity. In addition, if the object itself can be vibrated, or the camera can be vibrated, or the captured image can be vibrated, and the edge information can be generated based on the color information, the visual device can be expected even when the object is still. Standard (CNS) A4 specification (210 x 297 mm) (Please read the notes on the back before filling out this page) • Binding ---- Printed by the Intellectual Property Bureau Employee Consumer Cooperatives of the Ministry of Economic Affairs 540010-: ______B7 V. Invention Description (≫) Member of the Ministry of Economic Affairs, Intellectual Property Bureau 4 The number of objects can also be calculated if the cooperative has stopped printing. Here, it is assumed that there is the aforementioned vision device capable of counting the number of objects captured by the camera. When the camera system is stationary, the vision device can often count the number of stationary objects. However, when the object is moving, the vision device can only count the number of moving objects while the camera is shooting the moving object. Of course, if the position of stationary and moving objects can be specified in advance like cells in a petri dish, it is certainly not a problem, but if you are calculating people or animals moving in indoor or outdoor spaces, or because you ca n’t completely cover the entire indoor or outdoor space, Shooting into the camera's picture angle, or because the distance from the camera makes people or animals larger or smaller in the image, the use of the visual device will be limited if the camera is fixed. Furthermore, since visual devices must distinguish interior decoration from people or animals, a huge amount of calculation is required to identify objects. With the above in mind, the vision device that is expected is this. If the mobile camera can search for specific objects such as people or animals from indoor or outdoor spaces and only photograph those objects, and the visual device can adjust the magnification of the camera 'so that these objects become the appropriate size in the image, The visual device can easily identify these objects, and can also count the number of moving objects such as people or animals that cannot be specified in advance. Of course, when a person or animal is lying almost motionless, the visual device can also distinguish the person or animal from other stationary objects to calculate the number of people and animals. Therefore, the object of the present invention is to calculate the number of moving objects or all objects at high speed based on selectively selecting edge information generated by moving objects or any one of all objects in a moving image. In addition, the other 5 paper sizes are in accordance with Chinese National Standard (CNS) A4 (210 X 297 mm). Please read it first.
% Z ί裝 頁 I% Z ί page I
訂 • I I IOrder • I I I
540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（十） 一目的在於，自動態圖像中計算移動物體及靜止物體的數 量，以高速地計算動態圖像中移動物體及靜止物體的比例 。本案再一目的在於，藉搜索能以移動式攝像機拍攝之範 圍內所存在之移動物體及靜止物體，高速地計算移動物體 及靜止物體的數量。 [發明之揭示] 申請專利範圍第1項之發明，係一種視覺裝置，針對 動態圖像中之移動物體，，其具有： 取得前述動態圖像之框圖像的機構； 將前述框圖像作爲數位圖像依序加以記憶的機構； 自前述數位圖像生成移動物體粗邊緣資訊圖像的機構 使用前述數位圖像、自前述移動物體粗邊緣資訊圖像 生成移動物體形成邊緣資訊圖像的機構； 檢測根據前述移動物體形成邊緣資訊圖像所區分出之 移動物體領域位置及大小的機構； 計算移動物體領域數量的機構；以及 保持前述移動物體領域數量的機構。 取得前述框圖像的前述機構，在當前述動態圖像爲類 比信號時，係將前述框圖像轉換爲數位信號，作爲前述數 位圖像。在當前述動態圖像爲前述數位信號時，若已被壓 縮的話即解開壓縮，若未被壓縮的話即原封不動地加以輸 入，之後，自前述動態圖像中切割出前述框圖像，作爲前 述數位圖像。將前述框圖像作爲前述數位圖像依序記憶的 6 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） -------------裝--------訂--------- (請先閱讀背面之注意事項再填寫本頁) 540010 經濟部智慧財產局員工消費合作社印製 A7 五、發明說明（Κ) 前述機構，係保持二次元相位關係，將前述所有框圖像之 像素記憶於記憶體中。自前述數位圖像生成前述移動物體 粗邊緣資訊圖像的前述機構，係對前述數位圖像之各像素 生成移動物體的邊緣資訊。使用前述數位圖像’根據前述 移動物體粗邊緣資訊圖像，生成前述移動物體形成邊緣資 訊圖像的前述機構，係使用前述數位圖像，將前述移動物 體的前述邊緣資訊形成爲更正確且明瞭的前述邊緣資訊。 檢測根據前述移動物體形成邊緣資訊圖像所區分出之前述 移動物體領域之前述位置及前述大小的前述機構’係將位 於前述移動物體領域之交界處的前述邊緣資訊的重心位置 及位於前述移動物體領域之交界處的前述邊緣資訊數用一 個像素加以代表。計算前述移動物體領域數量的前述機構 ，係計算代表前述移動物體領域的前述像素的數量。保持 前述移動物體領域數量的前述機構，係將代表前述移動物 體領域之前述像素之數量’以2之餘數表現或浮動小數點 表現等輸出端所須之形式加以輸出。前述各機構能分別並 列動作，又’生成前述移動物體粗邊緣圖像的前述機構、 生成前述移動物體形成邊緣資訊圖像的前述機構、以及檢 測前述移動物體領域之前述位置及前述大小的前述機構， 亦能在各自的內部’各像素並列動作。此外，該等前述機 構，全部皆係透過局部處理加以實現。因此，前述視覺裝 置不至受前述移動物體之形狀或大小的影響，能高速地計 算前述移動物體領域數量。又，與根據顏色資訊的分類等 抽出前述移動物體領域之情形相較，由於前述邊緣資訊不 7 本紙張尺度適用中國國家標準（CNS)A4規格（210 x 297公复了 -----1---丨丨 裝--------訂--------- (請先閱讀背面之注意事項再填寫本頁) 540010 經濟部智慧財產局員工消費合作社印製 Λ7 B7 五、發明說明（& ) 易受照明等的影響，且對透明的前述移動體亦不進行染色 ，而利用折射或反射生成前述邊緣資訊，故前述視覺裝置 的應用範圍很廣。因此，能適當地解決關於前述移動物體 計算的各種問題。 申請專利範圍第2項之發明，係一種視覺裝置，針對 動態圖像中之移動物體，其具有： 取得前述動態圖像之框圖像的機構； 將前述框圖像作爲數位圖像依序加以記憶的機構； 自前述數位圖像生成移動物體粗邊緣資訊圖像的機構 使用前述數位圖像、自前述移動物體粗邊緣資訊圖像 生成移動物體形成邊緣資訊圖像的機構； 使用前述移動物體形成邊緣資訊圖像自背景分離移動 物體領域的機構； 檢測前述移動物體領域位置及大小的機構； 計算移動物體領域數的機構；以及 保持前述移動物體領域數的機構。 本發明，係在申請專利範圍第1項之發明中，增加使 用前述移動物體形成邊緣資訊圖像，自前述背景分離出前 述移動物體領域的前述機構。使用前述移動物體形成邊緣 資訊圖像，自前述背景分離出前述移動物體領域的前述機 構’係能將前述移動物體的前述邊緣資訊作爲交界，並將 包含於前述背景之像素與包含於前述移動物體領域之前述 像素’分類爲不同的組。同時，檢測前述移動物體領域之 8 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） ---r---*-----裝--------訂--------- (請先閱讀背面之注意事項再填寫本頁) 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（?） 前述位置及前述大小的前述機構’係將前述移動物體領域 之前述重心位置及前述移動物體領域所1包含的前述像素數 ，用一個前述像素加以代表。前述各機構能分別並列動作 ，又，生成前述移動物體粗邊緣圖像的前述機構、生成前 述移動物體形成邊緣資訊圖像的前述機構'以及檢' 測前述 移動物體領域之前述位置及前述大小的前述機構，亦能在 各自的內部，各像素並列動作。此外’該等前述機構’全 部皆係透過局部處理加以實現。因此’前述視覺裝置不至 受前述移動物體之形狀或大小的影響’能高速地計算前述 移動物體領域數量。又’與根據顏色資訊的分類等抽出前 述移動物體領域之情形相較，由於前述邊緣資訊不易受照 明等的影響，且對透明的前述移動體亦不進行染色，而利 用折射或反射生成前述邊緣資訊，故前述視覺裝置的應用 範圍很廣。因此，能適當地解決關於前述移動物體計算的 各種問題。 申請專利範圍第3項之發明，係如申請專利第1或2 項之視覺裝置，其具有使前述數位圖像振動的機構，據以 取代前述移動物體而對所有物體進行所有物體領域數量的 計算。當在申請專利範圍第1項之視覺裝置中，追加使前 述數位圖像振動的前述機構時，其作用如下。在當前述動 態圖像爲類比信號時，係將前述框圖像轉換爲數位信號， 作爲前述數位圖像。在當前述動態圖像爲前述數位信號時 ，右已被壓縮的g舌即解開壓縮，若未被壓縮的話即原封不 動地加以輸入，之後，自前述動態圖像中切割出前述框圖 9 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公复） --------*-----裝--------訂--------- (請先閱讀背面之注意事項再填寫本頁) 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明（（?） 像，作爲前述數位圖像。將前述框圖像作爲前述數位圖像 依序記憶的前述機構，係保持二次元相位關係，將前述所 有框圖像之像素記憶於記憶體中。自前述振動圖像生成前 述所有物體粗邊緣資訊圖像的前述機構，係以圖像單位或 像素單位使前述數位圖像上下左右振動’以生成前述振動 圖像。據此，前述振動圖像中之前述所有物體看起來宛如 在移動般。自前述數位圖像生成前述所有物體粗邊緣資訊 圖像的前述機構，係對前述數位圖像之各像素生成前述所 有物體的邊緣資訊。使用前述數位圖像’根據前述所有物 體粗邊緣資訊圖像，生成前述所有物體形成邊緣資訊圖像 的前述機構，係使用前述數位圖像’將前述所有物體的前 述邊緣資訊形成爲更正確且明瞭的前述邊緣資訊。檢測根 據前述所有物體形成邊緣資訊圖像所區分出之前述所有物 體領域之前述位置及前述大小的前述機構’係將位於前述 所有物體領域之交界處的前述邊緣資訊的重心位置及位於 前述所有物體領域之交界處的前述邊緣資訊數用一個像素 加以代表。計算前述所有物體領域數量的前述機構，係計 算代表前述所有物體領域的前述像素的數量。保持前述所 有物體領域數量的前述機構，係將代表前述所有物體領域 之前述像素之數量，以2之餘數表現或浮動小數點表現等 輸出端所須之形式加以輸出。前述各機構能分別並列動作 ，又，生成前述所有物體粗邊緣圖像的前述機構、生成前 述所有物體形成邊緣資訊圖像的前述機構、以及檢測前述 所有物體領域之前述位置及前述大小的前述機構，亦能在 10 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） --I---------裝--------訂--------- (請先閱讀背面之注意事項再填寫本頁) 540010 A7 B7 五、發明說明（°() 各自的內部，各像素並列動作。此外，該等前述機構，全 部皆係透過局部處理加以實現。因此，前述視覺裝置不至 受前述所有物體之形狀或大小的影響，能高速地計算前述 所有物體領域數量。又，與根據顏色資訊的分類等抽出前 述所有物體領域之情形相較，由於前述邊緣資訊不易受照 明等的影響，且對透明的前述所有物體亦不進行染色，而 利用折射或反射生成前述邊緣資訊，故前述視覺裝置的應 用範圍很廣。因此，能適當地解決關於前述所有物體計算 的各種問題。 另一方面，當在申請專利範圍第2項之視覺裝置中， 增追加使前述數位圖像振動的前述機構時，其作用如下。 使用前述所有物體形成邊緣資訊圖像，自前述背景分離出 前述所有物體領域的前述機構，係能將前述所有物體之前 述邊緣資訊作爲交界，並將前述背景所包含之前述像素與 前述所有物體領域所包含之前述像素加以分類爲不同的組 。同時，檢測前述所有物體領域之前述位置及前述大小的 前述機構，係將前述所有物體領域之重心位置及前述所有 物體領域所包含的前述像素數，以一個前述像素加以代表 。前述各機構能分別並列動作，又，生成前述所有物體粗 邊緣圖像的前述機構、生成前述所有物體形成邊緣資訊圖 像的前述機構、以及檢測前述所有物體領域之前述位置及 前述大小的前述機構，亦能在各自的內部，各像素並列動 作。此外，該等前述機構，全部皆係透過局部處理加以實 現。因此，前述視覺裝置不至受前述所有物體之形狀或大 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） (請先閱讀背面之注意事項再填寫本頁)540010 Printed by A7 B7, Consumer Cooperative of Intellectual Property Bureau of the Ministry of Economic Affairs 5. Description of the Invention (10) One purpose is to calculate the number of moving objects and stationary objects from moving images, and calculate the moving objects and stationary objects from moving images at high speed. The proportion of the object. Another purpose of this case is to calculate the number of moving objects and stationary objects at high speed by searching for moving objects and stationary objects that can be captured by a moving camera. [Disclosure of Invention] The invention in the first scope of the patent application is a visual device for moving objects in a moving image, which includes: a mechanism for obtaining the frame image of the foregoing dynamic image; and using the frame image as A mechanism for sequentially storing digital images; a mechanism for generating a rough object information image of a moving object from the digital image; a mechanism for generating an object information image using the digital image and a moving object from the coarse object information image of the moving object A mechanism for detecting the position and size of the moving object area distinguished according to the aforementioned moving object forming edge information image; a mechanism for calculating the number of moving object areas; and a mechanism for maintaining the aforementioned number of moving object areas. The mechanism for obtaining the frame image, when the moving image is an analog signal, converts the frame image into a digital signal as the digital image. When the aforementioned moving image is the aforementioned digital signal, if it has been compressed, it is decompressed, and if it is not compressed, it is inputted as it is. Then, the aforementioned frame image is cut out from the aforementioned moving image as the The aforementioned digital image. 6 paper sizes in which the aforementioned frame image is sequentially memorized as the aforementioned digital image are applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) ------------- install- ------ Order --------- (Please read the precautions on the back before filling out this page) 540010 Printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 V. Description of the Invention (K) Is to maintain the phase relationship of the second element, and store the pixels of all the frame images in the memory. The mechanism for generating the rough object information image of the moving object from the digital image is to generate edge information of a moving object for each pixel of the digital image. Using the aforementioned digital image 'The aforementioned mechanism for generating an edge information image of the moving object based on the rough edge information image of the moving object uses the aforementioned digital image to form the edge information of the moving object more accurately and clearly The aforementioned edge information for. Detecting the aforementioned position of the aforementioned moving object field and the aforementioned size, which are distinguished according to the aforementioned moving object formation edge information image, are the position of the center of gravity of the aforementioned edge information at the boundary of the aforementioned moving object field and the aforementioned moving object The aforementioned edge information number at the boundary of the field is represented by one pixel. The aforementioned mechanism for calculating the number of the aforementioned moving object fields is to calculate the number of the aforementioned pixels representing the aforementioned moving object fields. The aforementioned mechanism which maintains the aforementioned number of moving object fields is to output the number of the aforementioned pixels representing the aforementioned moving object field 'in the form required by the output terminal such as a remainder of two or a floating-point representation. The aforementioned mechanisms can operate in parallel, respectively, and the aforementioned mechanism that generates the rough edge image of the moving object, the aforementioned mechanism that generates the edge information image of the moving object, and the aforementioned mechanism that detects the position and size of the moving object field Also, each pixel can operate side by side within its own interior. In addition, all of the aforementioned institutions are realized through local processing. Therefore, the visual device is not affected by the shape or size of the moving object, and can calculate the number of moving object areas at high speed. In addition, compared with the case where the aforementioned moving object field is extracted based on the classification of color information, etc., because the aforementioned edge information is not equal to 7 paper sizes, the Chinese National Standard (CNS) A4 specification is applicable (210 x 297) ----- 1 --- 丨 丨 Installation -------- Order --------- (Please read the precautions on the back before filling out this page) 540010 Printed by the Consumer Cooperative of Intellectual Property Bureau of the Ministry of Economic Affairs Λ7 B7 V. Explanation of the invention (&) It is susceptible to the effects of lighting, etc., and does not dye the transparent moving body, but uses refraction or reflection to generate the aforementioned edge information, so the application scope of the aforementioned visual device is very wide. Therefore, it can Appropriately solve various problems concerning the calculation of the aforementioned moving objects. The invention in the second patent application scope is a visual device for moving objects in a moving image, which has: a mechanism for obtaining the frame image of the moving image; A mechanism for sequentially storing the frame image as a digital image; a mechanism for generating a rough edge information image of a moving object from the digital image uses the digital image and a rough edge information image from the moving object A mechanism that generates a moving object to form an edge information image; a mechanism that uses the aforementioned moving object to form an edge information image to separate the moving object field from the background; a mechanism that detects the position and size of the moving object field; a mechanism that counts the number of moving object fields; and The number of mechanisms in the aforementioned moving object field. The present invention relates to the invention in item 1 of the scope of patent application, which uses the aforementioned moving object to form an edge information image, and separates the aforementioned mechanism in the aforementioned moving object field from the background. Using the aforementioned movement The object forms an edge information image, and the aforementioned mechanism that separates the aforementioned moving object field from the aforementioned background is capable of taking the aforementioned edge information of the moving object as a boundary, and combining the pixels included in the aforementioned background and the aforementioned included in the aforementioned moving object field. The pixels are classified into different groups. At the same time, 8 paper sizes of the aforementioned moving objects are detected. The paper size applies to the Chinese National Standard (CNS) A4 (210 X 297 mm) --- r --- * ----- install -------- Order --------- (Please read the notes on the back before filling this page) 540010 Printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7 V. Description of the Invention (?) The aforementioned position and the aforementioned size of the aforementioned mechanism 'are the aforementioned position of the center of gravity of the aforementioned moving object field and the aforementioned number of pixels included in the aforementioned moving object field 1 It is represented by one of the aforementioned pixels. Each of the aforementioned mechanisms can operate in parallel, and the aforementioned mechanism that generates the rough edge image of the moving object, the aforementioned mechanism that generates the edge information image of the moving object, and the detection of the moving object The aforementioned positions in the field and the aforementioned mechanism of the aforementioned size can also be operated in parallel within each pixel. In addition, all of these aforementioned mechanisms are realized through local processing. Therefore, 'the aforementioned visual device is not affected by the shape or size of the aforementioned moving object', it is possible to calculate the number of the aforementioned moving object fields at high speed. Compared with the case where the aforementioned moving object area is extracted based on the classification of color information, etc., because the aforementioned edge information is not easily affected by lighting and the like, and the aforementioned transparent moving body is not dyed, the aforementioned edges are generated by refraction or reflection. Information, so the scope of application of the aforementioned visual device is very wide. Therefore, various problems concerning the aforementioned calculation of moving objects can be appropriately solved. The invention claimed in item 3 of the scope of patent application is a visual device such as the item 1 or 2 of patent application, which has a mechanism for vibrating the aforementioned digital image, and replaces the aforementioned moving object to calculate the number of all object fields for all objects . When the aforementioned mechanism for vibrating the aforementioned digital image is added to the visual device of the first patent application scope, the effect is as follows. When the dynamic image is an analog signal, the frame image is converted into a digital signal as the digital image. When the aforementioned moving image is the aforementioned digital signal, the compressed right tongue of the g is uncompressed, and if not compressed, it is inputted as it is, and then the aforementioned block diagram 9 is cut out from the aforementioned moving image. This paper size applies to China National Standard (CNS) A4 specification (210 X 297 public copy) -------- * ----- installation -------- order ------- -(Please read the notes on the back before filling this page) Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 540010 A7 B7 V. Description of the invention ((?) Image as the aforementioned digital image. Use the aforementioned frame image as The aforementioned mechanism for sequentially storing the aforementioned digital images maintains a two-dimensional phase relationship and stores the pixels of all the frame images in the memory. The aforementioned mechanism for generating the rough edge information images of all the objects from the aforementioned vibration image, The digital image is vibrated up and down, left and right in image units or pixel units to generate the aforementioned vibration image. Accordingly, all the objects in the aforementioned vibration image appear to be moving. The aforementioned image is generated from the digital image Infographic of rough edges of all objects The aforementioned mechanism of is to generate the edge information of all the aforementioned objects for each pixel of the aforementioned digital image. Using the aforementioned digital image 'generates the aforementioned information of all the objects to form an edge information image based on the aforementioned rough object information image of all the objects, The aforementioned digital image is used to form the aforementioned edge information of all the aforementioned objects into the aforementioned edge information which is more accurate and clear. Detects the aforementioned positions and the aforementioned sizes of all the aforementioned object areas which are distinguished according to the aforementioned formation of the aforementioned edge information image of all the objects The aforementioned mechanism of is a pixel representing the position of the center of gravity of the aforementioned edge information located at the boundary of all the aforementioned object domains and the number of aforementioned edge information located at the boundary of all the aforementioned object domains. , Is to calculate the number of the aforementioned pixels representing all the aforementioned object fields. The aforementioned mechanism which maintains the aforementioned number of the aforementioned subject fields, is to represent the number of the aforementioned pixels in all the aforementioned object fields, represented by a remainder of 2 or a floating-point representation, etc. Required The aforementioned mechanisms can be operated in parallel, and the aforementioned mechanism that generates the rough edge image of all the objects, the aforementioned mechanism that generates the edge information image of all the objects, and the aforementioned positions and the aforementioned detections of all the object areas The size of the aforementioned institutions can also apply the Chinese National Standard (CNS) A4 specification (210 X 297 mm) at 10 paper sizes. --------- (Please read the precautions on the back before filling out this page) 540010 A7 B7 V. Description of the invention (° () Each inside, each pixel operates side by side. In addition, all of the aforementioned institutions, all All are achieved through local processing. Therefore, the aforementioned visual device is not affected by the shape or size of all the aforementioned objects, and can calculate the number of all aforementioned object fields at high speed. In addition, compared with the case where all the aforementioned object areas are extracted based on the classification of color information, etc., because the aforementioned edge information is not easily affected by lighting, etc., and all the aforementioned transparent objects are not dyed, the aforementioned edges are generated by refraction or reflection. Information, so the scope of application of the aforementioned visual device is very wide. Therefore, it is possible to appropriately solve various problems concerning the calculation of all the aforementioned objects. On the other hand, when the visual device of the second patent application scope is added with the aforementioned mechanism for vibrating the digital image, the effect is as follows. Using the aforementioned all objects to form an edge information image, and separating the aforementioned mechanism of all the aforementioned object fields from the aforementioned background, can use the aforementioned edge information of all the aforementioned objects as a boundary, and the aforementioned pixels included in the aforementioned background and the aforementioned all object fields The aforementioned pixels included are classified into different groups. At the same time, the aforementioned mechanism that detects the aforementioned position and the aforementioned size of all the object fields is represented by one aforementioned pixel by the position of the center of gravity of the aforementioned object fields and the aforementioned number of pixels contained in the aforementioned object fields. Each of the aforementioned mechanisms can operate in parallel, and the aforementioned mechanism for generating a rough edge image of all the objects, the aforementioned mechanism for generating an edge information image of all the aforementioned objects, and the aforementioned mechanism for detecting the aforementioned position and the aforementioned size of all the aforementioned object fields , Can also operate in parallel within each pixel. In addition, all of the aforementioned institutions are realized through partial processing. Therefore, the aforementioned visual devices are not subject to the shape or size of all the aforementioned objects. The paper size is subject to the Chinese National Standard (CNS) A4 (210 X 297 mm) (Please read the precautions on the back before filling out this page)
裝— II訂·! I 經濟部智慧財產局員工消費合作社印製 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明（/ 小的影響，能高速地計算前述所有物體領域數量。又，與 根據顏色資訊的分類等抽出前述所有物體領域之情形相較 ，由於前述邊緣資訊不易受照明等的影響，且對透明的前 述所有物體亦不進行染色，而利用折射或反射生成前述邊 緣資訊，故前述視覺裝置的應用範圍很廣。因此，能適當 地解決關於前述所有物體計算的各種問題。 當組合申請專利範圍第1項之視覺裝置，與在申請專 利範圍第1項之視覺裝置增加使前述數位圖像振動之前述 機構的申請專利範圍第3項之視覺裝置時，本發明可在申 請專利範圍第1項之視覺裝置與申請專利範圍第3項之視 覺裝置間，共用自前述數位圖像生成前述移動物體粗邊緣 資訊圖像的前述機構，與自前述振動圖像生成前述所有物 體粗邊緣資訊圖像的前述機構；亦能共用使用前述數位圖 像、自前述移動物體粗邊緣資訊圖像生成前述移動物體形 成邊緣資訊圖像的前述機構，與使用前述數位圖像、自前 述所有物體粗邊緣資訊圖像生成前述所有物體形成邊緣資 訊圖像的前述機構；亦能共用使用前述移動物體形成邊緣 資訊圖像、檢測區分出之前述移動物體領域之前述位置及 前述大小的前述機構，與使用前述所有物體形成邊緣資訊 圖像、檢測區分出之前述所有物體領域之前述位置及前述 大小的前述機構；亦能共用計算前述移動物體領域數量的 前述機構，與計算前述所有物體領域數量的前述機構；亦 能共用保持前述移動物體領域數量的前述機構，與保持前 述所有物體領域數量的前述機構。若有硬體量的限制，亦 12 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） ---i---II —---丨裝--------訂--------- (請先閱讀背面之注意事項再填寫本頁) 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（/7) 可共用該等前述機構。當然，若不共用的話，即能縮短計 數時間。又，與根據顏色資訊的分類等抽出前述移動物體 領域及前述所有物體領域之情形相較，由於前述邊緣資訊 不易受照明等的影響，且對透明的前述移動物體領域及前 述所有物體亦不進行染色，而利用折射或反射生成前述邊 緣資訊，故前述視覺裝置的應用範圍很廣。因此，能適當 地解決關於前述移動物體領域及前述所有物體計算的各種 問題。 另一方面，當組合申請專利範圍第2項之視覺裝置， 與在申請專利範圍第2項之視覺裝置中增加使前述數位圖 像振動的前述機構之申請專利範圍第3項之視覺裝置時， 本發明可在申請專利範圍第2項之視覺裝置與申請專利範 圍第3項之視覺裝置間，共用自前述數位圖像、生成前述 移動物體粗邊緣資訊圖像的前述機構，與根據前述振動圖 像，生成前述所有物體粗邊緣資訊圖像的前述機構；亦能 共用使用前述數位圖像、自前述移動物體粗邊緣資訊圖像 生成前述移動物體形成邊緣資訊圖像的前述機構，與使用 前述數位圖像、自前述所有物體粗邊緣資訊圖像、生成前 述所有物體形成邊緣資訊圖像的前述機構；亦能共用使用 前述移動物體形成邊緣資訊圖像、自前述背景分離出前述 移動物體領域的前述機構，與使用前述所有物體形成邊緣 資訊圖像，自前述背景分離出前述所有物體領域的前述機 構；亦能共用檢測前述移動物體領域之前述位置及前述大 小的前述機構，與檢測前述所有物體領域之前述位置及前 13 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） (請先閱讀背面之注意事項再填寫本頁) >裝--------訂--------- 540010 經濟部智慧財產局員工消費合作杜印製 A7 B7 五、發明說明（丨i) 述大小的前述機構；亦能共用計算前述移動物體領域數的 前述機構，與計算前述所有物體領域數的前述機構；亦能 共用保持前述移動物體領域數的前述機構，與保持前述所 有物體領域數的前述機構。若有硬體量的限制，亦可共用 該等前述機構。當然，若不共用的話，即能縮短計數時間 。又，與根據顏色資訊的分類等抽出前述移動物體領域及 前述所有物體領域之情形相較，由於前述邊緣資訊不易受 照明等的影響，且對透明的前述移動物體領域及前述所有 物體亦不進行染色，而利用折射或反射生成前述邊緣資訊 ，故前述視覺裝置的應用範圍很廣。因此，能適當地解決 關於前述移動物體領域及前述所有物體計算的各種問題。 進一步的，當組合申請專利範圍第1項之視覺裝置， 與在申請專利範圍第1項之視覺裝置中增追加使前述數位 圖像振動之前述機構的申請專利範圍第3項之視覺裝置的 視覺裝置，或組合申請專利範圍第2項之視覺裝置，與在 申請專利範圍第2項之視覺裝置中增追加使前述數位圖像 振動之前述機構之申請專利範圍第3項之視覺裝置的視覺 裝置，可具備有：保持前述移動物體領域數量、與自前述 所有物體領域數量減去前述移動物體領域數量之靜止物體 領域數量的機能，或保持前述移動物體比例、與自前述所 有物體去除前述移動物體之靜止物體比例的機能。爲前者 時，本發明之保持前述所有物體領域數的前述機構，能自 前述所有物體領域數量減去前述移動物體領域數量，以求 出前述靜止物體領域數量。是以，保持前述所有物體領域 14 本紙張尺度適用中國國家標準（CNS)A4規格（21G X 297公髮） ---r----------裝--------訂--------- 華 (請先閱讀背面之注意事項再填寫本頁) 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（t1)) 數的前述機構，可視需要輸出前述靜止物體領域數量以取 代輸出前述所有物體領域數量。若爲後者時，本發明之保 持前述移動物體領域數的前述機構，能將前述移動物體領 域數量除以前述所有物體領域數量，即能求出前述移動物 體之比例。是以，保持前述移動物體領域數的前述機構， 可視需要輸出前述移動物體領域的比例以取代輸出前述移 動物體領域數量。另外，保持前述所有物體領域數的前述 機構，由於可自前述所有物體領域數量減去前述移動物體 領域數量，即能求出前述靜止物體領域數量，因此將前述 靜止物體領域數量除以前述所有物體領域數量，即能求出 前述靜止物體的比例。是以，保持前述所有物體領域數的 前述機構，可視需要輸出前述靜止物體領域的比例以取代 輸出則述所有物體領域數。 又，申請專利範圍第1、2及3項之視覺裝置中，使用 攝像機所拍攝的前述動態圖像，係使複數之前述框圖像連 續’以使前述框圖像中的前述移動物體看起來像是在移動 。因此，藉由使數位相機或掃描器等、於不同時刻或不同 場所作成的前述靜止圖像連續，即能使前述框圖像中的前 述移動物體看起來像是在移動。使用前述數位相機或前述 掃描器等以取代前述攝像機，即能在計算移動速度慢之前 述移動物體的前述移動物體領域數量時，容易地調整前述 R爭止圖像攝影時刻的間隔。另外，當計算在培養皿中細胞 之^部數量等不一定需要前述攝像機之情形時，由於能使 用问解像度之前述數位相機來取代前述攝像機，故能實現 ——L.---------•裝--------訂---------^9— (請先閱讀背面之注意事項再填寫本頁) 本紙張尺度顧㈣®^ii (CNS)A4^(210 297公釐） 540010 A7 B7 五、發明說明（/f) 便宜且精度高的視覺裝置。因此，能適當地解決有關前述 移動物體及前述所有物體計算的各種問題。 申請專利範圍第4項之發明，係一種視覺裝置，針對 使用移動式攝像機所拍攝之動態圖像中的任意物體，具有 取得前述動態圖像之框圖像的機構； 將前述框圖像作爲數位圖像、依序加以記憶的機構； 自前述數位圖像、生成任意物體粗邊緣資訊圖像的機 構； 檢測以前述任意物體粗邊緣資訊圖像所區分之任意物 體領域的位置及大小的機構； 將前述移動式攝像機之方向及倍率轉換爲環境座標位 置的機構； 將前述任意物體區域的前述位置及大小轉換爲前述環 境座標的前述位置之機構； 自複數之前述任意物體領域的前述環境座標之前述位 置中，選擇一個的機構； 輸入來自外部之控制指令的機構； 生成爲了使前述移動式攝像機振動之振動指令的機構 控制前述移動式攝像機至所應移動之前述環境座標之 前述位置的機構；以及 生成控制前述移動式攝像機之攝像指令的機構。 爲了以適當大小拍攝前述移動式攝像機所拍攝之前述 16 (請先閱讀背面之注意事項再填寫本頁) I 裝--------訂---------· 經濟部智慧財產局員工消費合作社印製 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（丨f) 任意物體，前述視覺裝置調整移動式攝像機的前述方向及 前述倍率。自前述動態圖像，生成前述任意物體的邊緣資 訊，使前述邊緣資訊退縮之一連串的處理’係全部以局部 處理加以實現。因此，前述視覺裝置不受前述任意物體之 形狀及大小影響，能高速地以適當之大小拍攝前述任意物 體。又，與根據顏色資訊的分類等抽出前述任意物體之領 域的情形相較，由於前述邊緣資訊不易受照明等的影饗’ 故前述視覺裝置的應用範圍很廣。因此，能適當地解決關 於前述任意移動物體之攝影時的各種問題。 申請專利範圍第5項之發明，係如申請專利範圍第4 項之視覺裝置，其具有： 使用前述數位圖像、自前述任意物體粗邊緣資訊圖# ，生成前述任意物體形成邊緣資訊圖像的機構； 使用前述任意物體形成邊緣資訊圖像，自背景分離出 前述任意物體領域的機構； 將前述任意物體領域規格化的機構； 保持任意物體規格化圖像的機構； 辨識前述任意物體規格化圖像的機構； 保持辨識結果的機構； 生成前述環境座標所顯示之環境地圖的機構； 保持前述環境地圖的機構； 在前述環境地圖中，推算前述任意物體之前述位置的 機構； 計算任意物體數量的機構； 17 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 丨丨l·---.-----•裝--------訂--------- (請先閲讀背面之注意事項再填寫本真) 540010 A7 B7 五、發明說明（A) 保持前述任意物體數的機構；以及 對前述任意物體形成邊緣資訊圖像進行幾何解析的機 (請先閱讀背面之注意事項再填寫本頁) 構。 經濟部智慧財產局員工消費合作社印製 本發明之特徵如以下所述。第一，自前述動態圖像生 成前述任意物體之前述邊緣資訊，使用前述邊緣資訊，將 自前述背景分離出之前述任意物體領域加以規格化的一連 串處理，係全部利用局部處理加以實現。因此，前述視覺 裝置不受前述任意物體之形狀及大小的影響，能高速地生 成前述任意物體規格化圖像，故前述視覺裝置的應用範圍 很廣。因此，能適當地解決有關前述任意物體之的抽出及 規格化的各種問題。第二，自前述動態圖像生成前述任意 物體的前述邊緣資訊，使用前述邊緣資訊將自前述背景分 離出之前述任意物體領域加以規格化之一連串的處理，係 全部利用局部處理加以實現。因此，前述視覺裝置不受前 述任意物體之形狀及大小的影響，能高速地生成前述任意 物體規格化圖像。而且，由於前述任意物體規格化圖像不 包含前述背景，故辨識前述任意物體規格化圖像的方法， 與前述背景所包圍之前述任意物體領域的情形相較，不易 受前述背景或前述任意物體領域之前述位置及大小等的影 響，故前述視覺裝置的應用範圍很廣。因此，能適當地解 決有關前述任意物體之辨識上的各種問題。第三，由於係 自前述任意物體規格化圖像之辨識結果與前述移動式攝像 機之前述方向及前述倍率，作成顯示前述任意物體之種類 與前述環境座標中之位置的環境資料，因此前述視覺裝置 18 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明) 能作成前述環境地圖（前述環境資料的集合）。由於前述環 境地圖包含一定期間之前述環境資料，因此能進行前述任 意物體之前述各種類的分布與移動狀況之記錄，前述視覺 裝置的應用範圍很廣。因此，能適當地解決有關前述任意 物體之記錄的各種問題。第四，若使用前述環境地圖，即 能求出尙未辨識的前述任意物體之前述環境座標中之前述 位置。又，由於前述移動式攝像機能推算在能攝影之範圍 中所存在之前述任意物體的實際位置，故前述視覺裝置的 應用範圍很廣。因此，能適當地解決有關前述任意物體搜 索的各種問題。第五，由於自前述環境地圖所包含之前述 環境資料中，抽出顯示特定之前述辨識結果者加以計算， 前述視覺裝置即能計算前述任意物體中屬於特定之前述種 類的數量，故前述視覺裝置的應用範圍很廣。因此，能適 當地解決有關前述任意物體計算的各種問題。第六，由於 能藉解析前述任意物體所生成之前述邊緣資訊的大致形狀 ，推算前述任意物體規格化圖像中之前述任意物體之前述 種類，因此辨識前述任意物體規格化圖像之前述機構，能 更高速且正確地辨識前述任意物體規格化圖像。此外，當 前述任意物體規格化圖像中之前述任意物體非爲辨識對象 時，由於辨識前述任意物體規格化圖像的前述機構能在生 成前述辨識結果前，生成前述環境地圖，故前述視覺裝置 的應用範圍很廣。因此，能適當地解決有關前述任意物體 的辨識、搜索及計算上的各種問題。 申請專利範圍第6項之發明，係一種視覺裝置，其特 19 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） ~ ' (請先閱讀背面之注意事項再填寫本頁) ·裝 tr---------· 540010 A7 B7 五、發明說明（/t) 徵在於，針對能使數位圖像振動的資料處理裝置中之各個 配置成網格狀的陣列運算單元，係具備： 將前述陣列運算單元初始化的機構； 若無應輸入之前述數位圖像時，即結束處理的機構； 輸入前述數位圖像之各頻帶像素値的機構； 使前述數位圖像之前述各頻帶像素値上下左右振動的 機構；以及 輸出振動圖像之各頻帶像素値的機構。 也就是說，此係爲了將前述陣列運算單元所提供的前 述數位圖像的振動機能以數位技術來加以實現之算法的組 裝形態。將前述陣列運算單元配置成網格狀，使鄰近之前 述陣列運算單元相互結合，待設定前述陣列運算單元的各 參數的初始値後，將前述數位圖像以像素單位適當地加以 輸入，依序進行從前述數位圖像之各頻帶像素値的振動， 到前述振動圖像之各頻帶像素値的輸出，反覆進行至前述 數位圖像不再輸入爲止。據此，能利用一般的處理器，容 易地進行前述參數之修正。 申請專利範圍第7項之發明，係一種視覺裝置，其特 徵在於，針對能自數位圖像生成粗邊緣資訊圖像的資料處 理裝置中之配置成網格狀之各個陣列運算單元，係具備： 將前述陣列運算單元初始化的機構； 若無應輸入之前述數位圖像時，即結束處理的機構； 輸入前述數位圖像之各頻帶像素値的機構； 將前述數位圖像之前述各頻帶像素値加以平滑化，以 20 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） (請先閱讀背面之注意事項再填寫本頁) -· a·» mbm I I _ μ·· I I mb I 蜃· 經濟部智慧財產局員工消費合作社印製 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明（β) 生成平滑化圖像之各頻帶像素値的機構； 取前述平滑化圖像之前述各頻帶像素値的對數，以生 成對數轉換圖像之各頻帶像素値的機構； 對前述對數轉換圖像之前述各頻帶像素値進行淸晰化 ，以生成淸晰化圖像之各頻帶像素値的機構； 自前述淸晰化圖像之前述各頻帶像素値，減去輸入1 個前淸晰化圖像之各頻帶像素値，以生成時間差圖像之各 頻帶像素値的機構； 將前述淸晰化圖像之前述各頻帶像素値轉換爲輸入1 個前淸晰化圖像之各頻帶像素値的機構； 算出前述時間差圖像之各頻帶像素値的拉普拉斯運算 碼(Laplacian)，以生成時間差分拉普拉斯運算碼圖像的各頻 帶像素値的機構； 抓出前述時間差拉普拉斯運算碼圖像之前述各頻帶像 素値的零點，以生成時間差零點圖像之各頻帶像素値的機 構； 求出前述時間差零點圖像之前述各頻帶像素値的最大 値，以生成最大値的時間差零點圖像之前述各頻帶像素値 的機構； 算出前述淸晰化圖像之前述各頻帶像素値的拉普拉斯 運算碼，以生成時間差拉普拉斯運算碼圖像之各頻帶像素 値的機構； 抽出拉普拉斯運算碼圖像之前述各頻帶像素値之前述 零點，以生成零點圖像之各頻帶像素値的機構； 21 表紙張尺度適用中國國家標準（CNS)A4規格^210 X 297公 —-—*--------------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 經濟部智慧財產局員工消費合作社印製 540010 A7 ------- B7 五、發明說明（>〇 ) 求出前述零點圖像之前述各頻帶像素値的最大値，以 生成最大値零點圖像之頻帶像素値的機構； 求出前述最大値零點圖像之前述頻帶像素値與前述最 大値時間差零點圖像之前述頻帶像素値中較大者，以生成 .混合零點圖像之頻帶像素値的機構； 去除前述混合零點圖像之孔，以生成孔除去混合零點 圖像之頻帶像素値的機構； 去除前述孔除去混合零點圖像之孤立點及孤立孔，以 生成雜訊除去混合零點圖像之頻帶像素値的機構； 反轉前述雜訊除去混合零點圖像之前述各頻帶像素値 ，以生成粗邊緣資訊圖像之頻帶像素値的機構；以及 輸出前述粗邊緣資訊圖像之前述頻帶像素値的機構。 也就是說，此係爲了將前述陣列運算單元所提供的前 述數位圖像的振動機能以數位技術來加以實現之演算法的 實際形態。將前述陣列運算單元配置成網格狀，使鄰近之 前述陣列運算單元相互結合，待設定前述陣列運算單元的 各參數的初始値後，將前述數位圖像以像素單位適當地加 以輸入，依序進行從前述數位圖像之平化，到前述粗邊緣 資訊圖像之各頻帶像素値的輸出，反覆進行至前述數位圖 像不再輸入爲止。據此，能利用一般的處理器，容易地進 行前述參數之修正。又，前述陣列運算單元，未必需要嚴 密進行等待位於附近之前述排列裝置所傳送來的各種前述 圖像之附近圖像的接收。此係因爲無法接收自位於附近之 前述陣列運算單兀之各種則述圖像的前述附近像素時，等 22 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） --->--- I ^--------------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 540010 A7 B7 五、發明說明（>l) (請先閱讀背面之注意事項再填寫本頁) 待接收之前述陣列運算單元能使用本身的前述頻帶像素値 加以取代的關係。此時，雖然前述陣列運算單元所生成之 各種前述圖像的像素中可能有若干的雜訊，但在本發明中 ，幾乎所有的雜訊都被前述各種機構加以吸收。藉使用本 身之前述頻帶像素値加以取代的機構，能同時且簡單地實 現邊緣處理與超時處理。 申請專利範圍第8項之發明，係一種視覺裝置，其特 徵在於，針對能自粗邊緣資訊圖像生成形成邊緣資訊圖像 的資料處理裝置中之配置成網格狀之各個陣列運算單元， 係具備： 將前述陣列運算單元初始化的機構； 若無應輸入之前述數位圖像時，即結束處理的機構； 輸入前述數位圖像之各頻帶像素値及前述粗邊緣資訊 圖像之各頻帶像素値的機構； 將前述數位圖像之各頻帶像素値及前述粗邊緣資訊圖 像之各頻帶像素値加以分離的機構； 將則述數位圖像之前述各頻帶像素値加以平滑化，以 生成平滑化圖像之各頻帶像素値的機構； 經濟部智慧財產局員工消費合作社印製 取前述平滑化圖像之前述各頻帶像素値的對數，以生 成對數轉換圖像之各頻帶像素値的機構； 對前述對數轉換圖像之前述各頻帶像素値進行淸晰化 ，以生成淸晰化圖像之各頻帶像素値的機構； 計算前述淸晰化圖像之前述各頻帶像素値的拉普拉斯 運算碼，以生成拉普拉斯運算碼圖像之各頻帶像素値的機 23 本紙張尺度適用中國國家標準（CNS)A4規格010 X 297公釐） 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明 構·’ 抽出前述拉普拉斯運算碼圖像之前述各頻帶像素値的 零點，以生成零點圖像之各頻帶像素値的機構的機構； 求出前述零點圖像之前述各頻帶像素値的最大値’以 生成最大値零點圖像之頻帶像素値的機構； 反轉前述最大値零點圖像之前述頻帶像素値’以生成 基礎邊緣資訊圖像之頻帶像素値的機構； 爲趨近基礎邊緣資訊圖像之前述頻帶像素値’將前述 粗邊緣資訊圖像之頻帶像素値加以整形的機構； 內插前述粗邊緣資訊圖像之前述頻帶像素値的線寬’ 以生成形成邊緣資訊圖像之頻帶像素値的機構；以及 輸出前述形成邊緣資訊圖像之前述頻帶像素値的機構 〇 也就是說’此係爲了將則述陣列運舁卓兀所提供的則 述形成邊緣資訊圖像的生成機能以數位技術來加以實現之 算法的組裝形態。將前述陣列運算單元配置成網格狀，使 鄰近之前述陣列運算單元相互結合，待設定前述陣列運算 單元的各參數的初始値後，將前述數位圖像及前述粗邊緣 資訊圖像以像素單位適當地加以輸入，依序進行從前述數 位圖像與前述粗邊緣資訊圖像之分離，到前述粗邊緣資訊 圖像之各頻帶像素値的輸出，反覆進行至前述數位圖像及 前述粗邊緣資訊圖像不再輸入爲止。據此，能利用一般的 處理器’容易地進行前述參數之修正。又，前述陣列運算 單元’未必需要嚴密進行等待位於附近之前述排列裝置所 24 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） —.—---------------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 540010Equipment — Order II! I Printed by the Employees ’Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs Printed by the Employees’ Cooperatives of the Ministry of Economic Affairs ’s Consumer Cooperatives printed 540010 A7 B7 V. Description of the invention (/ Small impact, can calculate the number of all the aforementioned object fields at high speed. In addition, compared with the case where all the aforementioned object areas are extracted based on the classification of color information, etc., because the aforementioned edge information is not easily affected by lighting, etc., and all the aforementioned transparent objects are not dyed, the aforementioned edges are generated by refraction or reflection. Information, so the scope of application of the aforementioned visual device is very wide. Therefore, it can properly solve various problems regarding the calculation of all the aforementioned objects. When combining the visual device of the first patent scope and the visual device of the first patent scope When the visual device of the patent application scope item 3 of the aforementioned mechanism that vibrates the aforementioned digital image is added, the present invention can be shared between the visual device of the patent scope item 1 and the visual device of the patent scope item 3, The aforementioned mechanism for generating a rough edge information image of a moving object from a digital image is the same as before The aforementioned mechanism that generates the rough edge information image of all the objects by the vibration image; the aforementioned mechanism that uses the digital image and generates the moving object to form the edge information image from the moving object rough edge information image, and the use of the foregoing Digital image, the aforementioned mechanism for generating the aforementioned edge information image for all the objects from the rough edge information image of all the aforementioned objects; it can also use the aforementioned moving object to form the edge information image, and detect the aforementioned position of the aforementioned moving object area which is distinguished And the aforementioned mechanism of the aforementioned size, and the aforementioned mechanism of forming the edge information image using all the aforementioned objects, detecting the aforementioned positions of the aforementioned all object fields and detecting the aforementioned dimensions; and the aforementioned mechanism for calculating the number of the aforementioned moving object fields, and The aforesaid organization that calculates the number of all the aforementioned object fields; it can also share the aforesaid organization that maintains the aforementioned number of moving object fields and the aforesaid mechanism that maintains the aforementioned number of all object fields. If there is a limit on the amount of hardware, this paper size is also applicable to China Standard (CNS) A4 Grid (210 X 297 mm) --- i --- II ----- 丨 installed -------- order --------- (Please read the precautions on the back before filling (This page) 540010 Printed by the Consumer Property Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7 V. Invention description (/ 7) These aforementioned institutions can be shared. Of course, if they are not shared, the counting time can be shortened. And, according to color information Compared with the situation of the aforementioned moving object domain and all the aforementioned object domains, the aforementioned edge information is not easily affected by lighting, etc., and the transparent aforementioned moving object domain and all the aforementioned objects are not dyed. Instead, refraction or The reflection generates the aforementioned edge information, so the application scope of the aforementioned visual device is wide. Therefore, various problems regarding the aforementioned moving object field and all the aforementioned object calculations can be appropriately solved. On the other hand, when the visual device of the second patent application scope is combined with the visual device of the second patent application scope, the visual device of the third patent application scope of the aforementioned mechanism that vibrates the aforementioned digital image is added, The present invention can share the aforementioned mechanism for generating a rough edge information image of the moving object from the aforementioned digital image between the visual device of the second patent application scope and the visual device of the third patent application scope, and the aforementioned vibration image. Image, the aforementioned mechanism for generating the rough edge information image of all the aforementioned objects; the aforementioned mechanism for generating the aforementioned moving object to form the edge information image using the aforementioned digital image, and generating the aforementioned moving object from the rough object information image of the aforementioned moving object, and using the aforementioned digital Image, the rough edge information image from all the aforementioned objects, the aforementioned mechanism that generates all the aforementioned objects to form an edge information image; it can also share the aforementioned information using the moving object to form an edge information image, and separate the aforementioned moving object domain from the background Mechanism, and use all the aforementioned objects to form edge information images, from the aforementioned back Separate the aforementioned institutions in all the aforementioned object fields; also be able to share the aforementioned institutions in the aforementioned moving object domain and the aforementioned size, and detect the aforementioned positions in the aforementioned all object domains and the first 13 paper standards applicable to Chinese National Standards (CNS) A4 specification (210 X 297 mm) (Please read the precautions on the back before filling out this page) > Packing -------- Order --------- 540010 Employees of Intellectual Property Bureau, Ministry of Economic Affairs Consumption Cooperation Du printed A7 B7 V. Description of the Invention (丨 i) The aforementioned mechanism of the size; the aforementioned mechanism for calculating the number of the aforementioned moving object fields, and the aforementioned mechanism for calculating the number of all the aforementioned object fields; and the aforementioned mechanism can also be shared and maintained The aforementioned mechanism of the number of moving object fields and the aforementioned mechanism of holding the number of all object fields. If there is a limit on the amount of hardware, these aforementioned institutions can also be shared. Of course, if it is not shared, it can shorten the counting time. In addition, compared with the case where the aforementioned moving object domain and all the aforementioned object domains are extracted based on the classification of color information, etc., the aforementioned edge information is not easily affected by lighting, etc. The aforementioned edge information is generated by refraction or reflection, so the application range of the aforementioned visual device is very wide. Therefore, it is possible to appropriately solve various problems regarding the aforementioned moving object field and all the aforementioned object calculations. Further, when the visual device of the first patent application scope is combined with the visual device of the first patent application scope, the visual device of the third patent application scope of the aforementioned mechanism that vibrates the aforementioned digital image is added to the visual device of the first patent application scope. Device, or a combination of the visual device of the second patent application scope, and the visual device of the second patent application scope, and a visual device of the third patent application scope of the aforementioned mechanism that vibrates the aforementioned digital image It can have the function of maintaining the number of the moving object fields and the number of stationary object fields minus the number of the moving object fields from the number of all the object fields, or maintaining the ratio of the moving objects and removing the moving objects from all the objects. The function of the proportion of stationary objects. In the former case, the aforementioned mechanism of the present invention holding the number of all object fields can subtract the number of moving object fields from the number of all object fields to obtain the number of stationary object fields. Therefore, keep all the above-mentioned object areas 14 This paper size is applicable to the Chinese National Standard (CNS) A4 specification (21G X 297 issued) --- r ---------- install ------- -Order --------- Hua (Please read the notes on the back before filling out this page) 540010 Printed by A7 B7, Consumer Cooperatives of the Intellectual Property Bureau, Ministry of Economic Affairs, V. Invention Description (t1)) If necessary, the foregoing number of the still object fields may be output instead of outputting the foregoing number of all object fields. In the latter case, the aforementioned mechanism of the present invention that maintains the number of moving object fields can divide the number of moving object fields by the number of all object fields to obtain the ratio of the moving object. Therefore, the mechanism that maintains the number of moving object fields may output the proportion of the moving object fields as needed instead of outputting the number of moving object fields. In addition, the aforementioned mechanism holding the number of all object fields can subtract the number of moving object fields from the number of all object fields to obtain the number of stationary object fields. Therefore, the number of the stationary object fields is divided by the number of all objects. The number of fields, that is, the ratio of the stationary object can be obtained. Therefore, the mechanism that maintains the number of all object fields may output the proportion of the still object fields as required instead of outputting the number of all object fields. Furthermore, in the visual devices of the scope of patent applications 1, 2 and 3, the use of the moving images captured by a camera is to continuously make a plurality of the frame images to make the moving objects in the frame images look like It seems to be moving. Therefore, by continuing the aforementioned still images created at different times or in different places by a digital camera or a scanner, it is possible to make the aforementioned moving object in the aforementioned frame image appear to be moving. When the digital camera, the scanner, or the like is used instead of the video camera, it is possible to easily adjust the interval of the R-continuing image capturing time when calculating the number of the moving object areas where the moving speed is slow before the moving object. In addition, when calculating the number of cells in a petri dish, such as when the aforementioned camera is not necessarily required, the aforementioned digital camera can be used instead of the aforementioned camera, so it can be achieved-L .------ --- • Installation -------- Order --------- ^ 9— (Please read the precautions on the back before filling this page) This paper size Gu㈣® ii (CNS) A4 ^ (210 297 mm) 540010 A7 B7 V. Description of the invention (/ f) Cheap and highly accurate vision device. Therefore, various problems concerning the calculation of the aforementioned moving objects and all the aforementioned objects can be appropriately solved. The invention in the fourth scope of the patent application is a visual device having a mechanism for acquiring a frame image of the foregoing dynamic image for any object in a moving image captured by a moving camera; the frame image is used as a digital An image and a mechanism for sequentially storing the image; a mechanism for generating a rough edge information image of an arbitrary object from the foregoing digital image; a mechanism for detecting the position and size of an arbitrary object area distinguished by the foregoing rough image of the arbitrary object information; A mechanism that converts the direction and magnification of the aforementioned mobile camera into the position of the environmental coordinate; a mechanism that converts the aforementioned position and size of the aforementioned arbitrary object area into the aforementioned position of the aforementioned environmental coordinate; A mechanism for selecting one of the aforementioned positions; a mechanism for inputting a control command from the outside; a mechanism for generating a vibration command for vibrating the mobile camera to control the mobile camera to the aforementioned position of the aforementioned environmental coordinates to be moved; And generating and controlling the aforementioned mobile camera The camera instruction body. In order to shoot the aforementioned 16 taken by the mobile camera in an appropriate size (please read the precautions on the back before filling out this page) I equipment -------- Order --------- · Ministry of Economic Affairs Printed by the Intellectual Property Bureau Staff Consumer Cooperatives This paper is printed in accordance with the Chinese National Standard (CNS) A4 (210 X 297 mm) 540010 Printed by the Intellectual Property Bureau Staff Consumer Cooperatives A7 B7 V. Description of the invention (丨 f) Any object The aforementioned visual device adjusts the aforementioned direction and the aforementioned magnification of the mobile camera. A series of processes of generating edge information of the arbitrary object from the dynamic image and shrinking the edge information are all implemented by local processing. Therefore, the aforementioned visual device is not affected by the shape and size of the aforementioned arbitrary object, and can photograph the aforementioned arbitrary object at an appropriate size at high speed. Furthermore, compared with the case where the area of the arbitrary object is extracted based on the classification of color information and the like, the edge information is not easily affected by lighting and the like, so the visual device has a wide application range. Therefore, it is possible to appropriately solve various problems concerning photography of the aforementioned arbitrary moving object. The invention in the fifth scope of the patent application is the visual device as in the fourth scope of the patent application, which includes: using the aforementioned digital image and the rough edge information map # of any arbitrary object to generate an edge information image of the aforementioned arbitrary object Mechanism; using the aforementioned arbitrary object to form an edge information image, to separate the aforementioned arbitrary object field from the background; mechanism to normalize the aforementioned arbitrary object field; mechanism to maintain the standardized image of the arbitrary object; to identify the aforementioned arbitrary object normalization map A mechanism that maintains the recognition result; a mechanism that generates the environmental map displayed by the aforementioned environmental coordinates; a mechanism that maintains the aforementioned environmental map; a mechanism that estimates the aforementioned position of the aforementioned arbitrary object in the aforementioned environmental map; Institution; 17 This paper size applies to China National Standard (CNS) A4 (210 X 297 mm) 丨 丨 l · ---.----- • Installation -------- Order ---- ----- (Please read the precautions on the back before filling in the true) 540010 A7 B7 V. Description of the invention (A) The mechanism to keep the number of the aforementioned arbitrary objects; Machine intended to form the edges of objects infographics geometric resolution (please read the back of the precautions to fill out this page) configuration. Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs The features of the present invention are as follows. First, the edge information of the arbitrary object is generated from the dynamic image, and a series of processes of normalizing the area of the arbitrary object separated from the background using the edge information are all realized by local processing. Therefore, the aforementioned visual device is not affected by the shape and size of the aforementioned arbitrary object, and can generate a normalized image of the aforementioned arbitrary object at high speed. Therefore, the aforementioned vision device has a wide application range. Therefore, various problems concerning the extraction and normalization of any of the aforementioned objects can be appropriately solved. Second, the edge information of the arbitrary object is generated from the dynamic image, and the edge information is used to normalize the arbitrary object area separated from the background to perform a series of processes, all of which are implemented using local processing. Therefore, the aforementioned visual device can generate the normalized image of the aforementioned arbitrary object at high speed without being affected by the shape and size of the aforementioned arbitrary object. Moreover, since the aforementioned normalized image of any object does not include the aforementioned background, the method of identifying the aforementioned normalized image of any object is less susceptible to the aforementioned background or the aforementioned arbitrary object than the situation of the aforementioned arbitrary object area surrounded by the aforementioned background. Due to the influence of the aforementioned position and size in the field, the application range of the aforementioned visual device is very wide. Therefore, various problems regarding the identification of any of the foregoing objects can be appropriately solved. Third, because the recognition results of the normalized image of the arbitrary object and the aforementioned direction and magnification of the mobile camera are used to create environmental data showing the type of the arbitrary object and the position in the aforementioned environmental coordinates, the aforementioned visual device 18 This paper size applies to China National Standard (CNS) A4 (210 X 297 mm) Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 540010 A7 B7 V. Description of the invention) The aforementioned environmental map (collection of the aforementioned environmental information) . Because the aforementioned environment map contains the aforementioned environmental data for a certain period of time, it is possible to record the aforementioned various types of distribution and movement of any object, and the application scope of the aforementioned visual device is very wide. Therefore, various problems concerning the recording of any of the foregoing objects can be appropriately solved. Fourth, if the aforementioned environmental map is used, the aforementioned position in the aforementioned environmental coordinates of any unidentified aforementioned arbitrary object can be obtained. In addition, since the mobile camera can estimate the actual position of the arbitrary object existing in the range that can be photographed, the application range of the visual device is very wide. Therefore, various problems related to the aforementioned arbitrary object search can be appropriately solved. Fifth, since the environmental data included in the environmental map is extracted and calculated by displaying the specific identification result, the visual device can calculate the number of the specific types among the arbitrary objects. The application range is very wide. Therefore, various problems related to the calculation of any of the foregoing objects can be appropriately solved. Sixth, because the approximate shape of the edge information generated by the arbitrary object can be analyzed to estimate the aforementioned type of the arbitrary object in the arbitrary object normalized image, the aforementioned mechanism for identifying the arbitrary object normalized image is identified, The normalized image of any of the aforementioned objects can be recognized more quickly and accurately. In addition, when the above-mentioned arbitrary object in the above-mentioned arbitrary object normalized image is not a recognition object, since the aforementioned mechanism that recognizes the above-mentioned normalized image of any arbitrary object can generate the aforementioned environment map before generating the aforementioned recognition result, the aforementioned visual device The application range is very wide. Therefore, various problems related to the identification, search, and calculation of any of the foregoing objects can be appropriately solved. The invention in the 6th scope of the patent application is a visual device, and its special 19 paper sizes are applicable to the Chinese National Standard (CNS) A4 (210 X 297 mm) ~ '(Please read the precautions on the back before filling this page ) · Install tr --------- · 540010 A7 B7 V. Description of the invention (/ t) The feature lies in that each of the data processing devices capable of vibrating a digital image is arranged in a grid-like array operation. The unit is provided with: a mechanism for initializing the aforementioned array operation unit; a mechanism for ending processing if the aforementioned digital image should not be input; a mechanism for inputting pixels of each band of the aforementioned digital image; The aforementioned mechanism for the pixels 各 in each frequency band to vibrate up, down, left and right; and the mechanism for outputting the pixels 各 in each frequency band of a vibration image. That is, this is an assembly form of an algorithm for realizing the digital image vibration function provided by the aforementioned array operation unit with digital technology. The array operation units are arranged in a grid shape, so that the adjacent array operation units are combined with each other. After the initial parameters of the parameters of the array operation units are set, the digital images are appropriately input in pixel units and sequentially The vibration is performed from the pixel 値 of each frequency band of the digital image to the output of the pixel 値 of each frequency band of the vibration image, until the digital image is no longer input. This makes it possible to easily modify the aforementioned parameters by using a general processor. The invention of claim 7 in the scope of patent application is a visual device, which is characterized in that, for each array operation unit arranged in a grid shape in a data processing device capable of generating a rough edge information image from a digital image, it has: A mechanism for initializing the aforementioned array operation unit; a mechanism for ending processing if there is no aforementioned digital image to be input; a mechanism for inputting pixels of each frequency band of the aforementioned digital image; Smoothed and applied Chinese National Standard (CNS) A4 specification (210 X 297 mm) at 20 paper sizes (Please read the precautions on the back before filling this page)-· a · »mbm II _ μ ·· II mb I 蜃 · Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs and printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs and printed by the Consumer Cooperatives of the Ministry of Economic Affairs. A mechanism for transforming the logarithm of the pixel 値 in each band of the image to generate a logarithmic conversion of the pixel 値 in each band of the image; Band sharpening mechanism for band pixels 淸 to generate sharpened images of each band pixel 减去 of the image; from the aforementioned sharpened image of each band pixel 値, subtracting each of the input 1 previously sharpened image Band pixel 値, a mechanism for generating each band pixel 値 of the time difference image; a mechanism for converting each band pixel 値 of the aforementioned sharpened image into each band pixel 値 of a previously sharpened image; calculation Laplacian of pixel bands 値 in each frequency band of the aforementioned time difference image to generate a mechanism of pixel bands 各 of each band in time differential Laplace opcode image; capture the aforementioned Laplacian code of time difference The zero point of the pixel 値 of each band in the image is used to generate a mechanism of the pixel 値 of each band of the time difference zero point image. The maximum value of the pixel 値 of each band of the time difference zero point image is obtained to generate a time difference zero point map of the maximum Like the aforementioned mechanism of pixel 値 in each frequency band; calculating the Laplacian operation code of the aforementioned pixel 値 in each frequency band to generate a time difference Laplacian operation code image The mechanism of the pixels in each frequency band; the mechanism of extracting the aforementioned zero points of the aforementioned pixels in each frequency band of the Laplacian opcode image to generate the pixels 各 of each frequency band in the zero point image; 21 The paper scale applies the Chinese National Standard (CNS) A4 Specifications ^ 210 X 297 Male —-— * ------------ Order --------- Line (Please read the precautions on the back before filling this page) Economy Printed by the Consumers' Cooperative of the Ministry of Intellectual Property Bureau 540010 A7 ------- B7 V. Description of the invention (> 〇) Find the maximum value of the pixel 値 of each frequency band of the aforementioned zero point image to generate the maximum zero point map The mechanism of the band pixel 値 of the image; the larger of the band pixel 値 of the aforementioned maximum zero point image and the band pixel 値 of the aforementioned maximum time difference zero point image is generated to generate a band pixel of the mixed zero point image 値Mechanism; removing the holes of the mixed zero point image to generate holes to remove the band pixels of the mixed zero point image; removing the holes to remove the isolated points and isolated holes of the mixed zero point image to generate the noise removing mixed zero point map Image-band pixel 値 mechanism; Turn the noise removing the mixed zero-point image of each band-pixel Zhi, means to generate a band-pixel of the rough edge-information image Zhi; and an output of the band-pixel means the rough edge-information image of the Zhi. In other words, this is an actual form of an algorithm for realizing the digital image vibration function provided by the aforementioned array operation unit with digital technology. The array operation units are arranged in a grid shape, so that the adjacent array operation units are combined with each other. After the initial parameters of the parameters of the array operation units are set, the digital images are appropriately input in pixel units and sequentially The flattening of the digital image to the output of pixels 値 in each frequency band of the rough edge information image is performed repeatedly until the digital image is no longer input. This makes it possible to easily modify the aforementioned parameters using a general processor. In addition, the array operation unit does not necessarily need to wait closely for the reception of nearby images of the various images transmitted from the array device located nearby. This is because when it is not possible to receive the aforementioned nearby pixels of various images described by the aforementioned array operation unit located nearby, etc. 22 paper sizes are applicable to China National Standard (CNS) A4 specifications (210 X 297 mm) --- > --- I ^ -------------- Order --------- line (Please read the precautions on the back before filling this page) 540010 A7 B7 V. Description of the Invention (> l) (Please read the notes on the back before filling this page) The aforementioned array operation unit to be received can be replaced by the aforementioned band pixel 値. At this time, although there may be some noise in the pixels of the various images generated by the array operation unit, in the present invention, almost all noises are absorbed by the aforementioned various mechanisms. By using the above-mentioned band pixel unit to replace it, the edge processing and timeout processing can be implemented simultaneously and simply. The invention in the eighth aspect of the patent application is a visual device characterized in that each array operation unit arranged in a grid shape in a data processing device capable of generating an edge information image from a coarse edge information image is a system It is provided with: a mechanism for initializing the aforementioned array operation unit; a mechanism for ending the processing if there is no aforementioned digital image to be input; inputting each band pixel of the aforementioned digital image 图像 and each band pixel of the aforementioned rough edge information image 边缘A mechanism that separates each band pixel 値 of the digital image and each band pixel 粗 of the rough edge information image; smoothes the band pixel 各 of the digital image to generate a smoothing The mechanism of pixel bands in each band of the image; the consumer cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs prints the logarithm of the pixel bands in each band of the aforementioned smoothed image to generate a logarithmic conversion image of the pixel bands in each band; Pixels of each band of the aforementioned log-transformed image are sharpened to generate pixels of each band of the sharpened image The machine that calculates the Laplacian opcodes of the aforementioned pixels in each frequency band of the sharpened image to generate the Laplacian opcodes of the pixels in each band of the image. 23 This paper applies Chinese national standards ( CNS) A4 size 010 X 297 mm) Printed by the Consumers 'Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 540010 A7 B7 V. Description of the invention ·' Extract the zero points of the pixels of each band in the aforementioned Laplacian opcode image to A mechanism for generating a pixel 各 of each frequency band of the zero point image; a mechanism for obtaining a maximum 値 ′ of the frequency band pixels 各 of the aforementioned zero point image to generate a maximum 値 band pixel 値 of the zero point image; inverting the aforementioned maximum zero point A mechanism for generating the aforementioned band pixels 値 of the image to generate the band pixels 基础 of the basic edge information image; To approach the aforementioned band pixels 値 of the basic edge information image, the band pixels 値 of the aforementioned rough edge information image are shaped A mechanism for interpolating the line widths of the aforementioned frequency band pixels 値 of the rough edge information image to generate the frequency band pixels 形成 forming the edge information image; and The above-mentioned mechanism for forming the aforementioned band pixels of the edge information image is to be described. That is, 'the purpose is to realize the generation function of the edge information image provided by the array array using the digital technology. The assembly form of the algorithm. The array operation unit is arranged in a grid shape, so that the adjacent array operation units are combined with each other. After setting the initial parameters of the parameters of the array operation unit, the digital image and the rough edge information image are set in pixel units. Appropriate input, sequentially separating from the digital image and the rough edge information image, to the output of each band pixel 値 of the rough edge information image, repeatedly to the digital image and the rough edge information The image is no longer input. This makes it possible to easily modify the aforementioned parameters using a general processor '. In addition, the aforementioned array operation unit does not necessarily need to wait closely for the above-mentioned arrangement device located nearby. The paper size is applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) —.—-------- ------- Order --------- Line (Please read the precautions on the back before filling this page) 540010
五、發明說明（V7)) 經濟部智慧財產局員工消費合作社印製 傳送來的各種前述圖像之附近圖像的接收。此係因爲無法 接收自位於附近之前述陣列運算單元之各種前述圖像的前 述附近像素時，等待接收之前述陣列運算單元能使用本身 的前述頻帶像素値加以取代的關係。此時，雖然前述陣列 運算單兀所生成之各種前述圖像的像素中可能有若干的雜 訊，但在本發明中，幾乎所有的雜訊都被前述各種機構^ 以吸收。藉使用本身之前述頻帶像素値加以取代的機構， 能同時且簡單地實現邊緣處理與超時處理。 ’ 申請專利範圍第9項之發明，係一種視覺裝置， 具特 徵在於，針對能檢測物體領域之位置及大小之機構的資半斗 處理裝置中之配置成網格狀之各個陣列運算單元，保 丨木具備 將前述陣列運算單元初始化的機構； 若無應輸入之前述數位圖像時，即結束處理的機_ ; 輸入前述粗邊緣資訊.圖像之各頻帶像素値的機梅； 將前述粗邊緣資訊圖像之頻帶像素値轉換爲重複 圖像之頻帶像素値的機構； 將自前述重複資訊圖像所計算的移動量圖像化爲 量圖像之頻帶像素値的機構； 將前述重複資訊圖像之前述頻帶像素値移動至前述& 動量圖像之前述頻帶像素値所示位置的機構； 將前述重複資訊圖像之前述頻帶像素値更新爲前述亀 複資訊圖像之移動源之前述頻帶像素値之合計的機構；以 及 25 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公爱） ---*---1------^^衣--------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 540010 Α7 Β7 五、發明說明 輸出前述重複資訊圖像之前述頻帶像素値的機構。 (請先閱讀背面之注意事項再填寫本頁) 經濟部智慧財產局員工消費合作社印製 也就是說，此係爲了將前述陣列運算單元所提供的前 述粗邊緣資訊圖像所指之顯示前述物體領域的前述位置及 前述大小之前述重複資訊圖像的生成機能以數位技術來加 以實現之算法的組裝形態。將前述陣列運算單元配置成網 格狀，使鄰近之前述陣列運算單元相互結合，待設定前述 陣列運算單元的各參數的初始値後，將前述數位圖像及前 述粗邊緣資訊圖像以像素單位適當地加以輸入，依序進行 從變換爲前述重複資訊圖像，到前述重複資訊圖像之各頻 帶像素値的輸出，反覆進行至前述數位圖像及前述粗邊緣 資訊圖像不再輸入爲止。據此，能利用一般的處理器，容 易地進行前述參數之修正。又，前述陣列運算單元，未必 需要嚴密進行等待位於附近之前述排列裝置所傳送來的各 種前述圖像之附近圖像的接收。此係因爲無法接收自位於 附近之前述陣列運算單元之各種前述圖像的前述附近像素 時，等待接收之前述陣列運算單元能使用相當於0之像素 値加以取代的關係。此時，雖然前述陣列運算單元所生成 之各種前述圖像的像素中可能有若干的雜訊，但在本發明 中，幾乎所有的雜訊都被前述各種機構加以吸收。藉使用 相當於0之前述頻帶像素値加以取代的機構，能同時且簡 單地實現邊緣處理與超時處理。 本發明，可輸入形成邊緣資訊圖像以取代前述粗邊緣 資訊圖像。爲了生成顯示前述形成邊緣資訊圖像所指之前 述物體領域之前述位置及前述大小的前述重複資訊圖像， 26 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（<) 將前述陣列運算單元配置成網格狀，使鄰近之前述陣列運 算單元相互結合，待設定前述陣列運算單元的各參數的初 始値後，將前述數位圖像及前述粗邊緣資訊圖像以像素單 位適當地加以輸入’依序進行從變換爲前述重複資訊圖像 ，到前述重複資訊圖像之各頻帶像素値的輸出，反覆進行 至前述數位圖像及前述粗邊緣資訊圖像不再輸入爲止。據 此，能利用一般的處理器，容易地進行前述參數之修正。 又，前述陣列運算單元’未必需要嚴密進行等待位於附近 之前述排列裝置所傳送來的各種前述圖像之附近圖像的接 收。此係因爲無法接收自位於附近之前述陣列運算單元之 各種前述圖像的前述附近像素時，等待接收之前述陣列運 算單元能使用相當於0之像素値加以取代的關係。此時， 雖然前述陣列運算單元所生成之各種前述圖像的像素中可 能有若干的雜訊’但在本發明中，幾乎所有的雜訊都被前 述各種機構加以吸收。藉使用相當於〇之前述轉帶像素値 加以取代的機構，能同時且簡單地實現邊緣處理與超時處 理。 又，本發明可輸入前述物體領域圖像以取代；前述粗邊 緣資訊圖像。爲了生成顯不則述形成邊緣資訊圖丨象所指之 前述物體領域之前述位置及前述大小的前述重複畜訊 ，將前述陣列運算單元配置成網格狀，使鄰近之前述陣列 運算單元相互結合，待設定前述陣列運算單元的各參數的 初始値後，將前述數位圖像及前述粗邊緣資訊圖像以像素 單位適當地加以輸入，依序進行從變換爲前述重複資訊圖 27 本紙張尺度適用中國國家標準(CNS)A4規格（210 X 297公f ^^—- ---,---,-----------I--訂---------線 (請先閱讀背面之注意事項再填寫本頁) 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明（vL ) 像，到前述重複資訊圖像之各頻帶像素値的輸出，反覆進 行至前述數位圖像及前述粗邊緣資訊圖像不再輸入爲止。 據此，能利用一般的處理器，容易地進行前述參數之修正 。又，前述陣列運算單元，未必需要嚴密進行等待位於附 近之前述排列裝置所傳送來的各種前述圖像之附近圖像的 接收。此係因爲無法接收自位於附近之前述陣列運算單元 之各種前述圖像的前述附近像素時，等待接收之前述陣列 運算單元能使用相當於0之像素値加以取代的關係。此時 ，雖然前述陣列運算單元所生成之各種前述圖像的像素中 可能有若干的雜訊，但在本發明中，幾乎所有的雜訊都被 前述各種機構加以吸收。藉使用相當於0之前述頻帶像素 値加以取代的機構，能同時且簡單地實現邊緣處理與超時 處理。 申請專利範圍第10項之發明，係一種視覺裝置，其特 徵在於，針對能檢測物體領域之位置及大小之機構的資料 處理裝置中之配置成網格狀之各個陣列運算單元，係具備 將前述陣列運算單元初始化的機構； 若無應輸入之前述數位圖像時，即結束處理的機構； 輸入前述粗邊緣資訊圖像之各頻帶像素値的機構； 輸入前述物體領域圖像之頻帶像素値及前述數位圖像 之各頻帶像素値的機構； 分離前述物體領域圖像之前述頻帶像素値與前述數位 圖像之前述各頻帶像素値，以生成更新物體領域圖像之頻 28 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐) ' ---*------------------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 540010V. Description of the invention (V7)) Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs. This is because when the aforementioned nearby pixels of various aforementioned images cannot be received from the aforementioned array computing unit located nearby, the aforementioned array computing unit waiting to be received can be replaced by the aforementioned band pixel 频带 of itself. At this time, although there may be some noise in the pixels of the various aforementioned images generated by the aforementioned array operation unit, in the present invention, almost all the noises are absorbed by the aforementioned various mechanisms ^. By using the aforementioned mechanism to replace the aforementioned band pixel unit, edge processing and timeout processing can be implemented simultaneously and simply. '' The invention in the ninth scope of the patent application is a visual device, which is characterized in that each array computing unit arranged in a grid shape in a semi-bucket processing device of a mechanism capable of detecting the position and size of an object field, protects丨 There is a mechanism for initializing the aforementioned array operation unit; if there is no aforementioned digital image that should be input, the processing is ended; input the aforementioned rough edge information. The image of each band of pixels in the image of the machine; A mechanism for converting band pixels 値 of an edge information image into band pixels 重复 of a repeated image; a mechanism for converting a movement amount calculated from the aforementioned repeated information image into band pixels 量 of a quantity image; A mechanism for moving the aforementioned band pixel 値 of the image to the position indicated by the aforementioned band pixel & of the & momentum image; updating the aforementioned band pixel 前述 of the repeating information image to the aforementioned moving source of the complex information image The organization of the total number of band pixels; and 25 paper sizes are applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 public love) --- * --- 1 ------ ^^ clothing -------- order --------- line (please read the precautions on the back before filling this page) 540010 Α7 Β7 V. Description of the invention Band pixel 値 mechanism. (Please read the notes on the back before filling out this page) Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs That is, this is to display the aforementioned objects indicated by the aforementioned rough edge information images provided by the aforementioned array computing unit The assembly position of the algorithm in which the aforementioned position of the field and the aforementioned generation of the repeated information image of the size can be realized by digital technology. The array operation unit is arranged in a grid shape, so that the adjacent array operation units are combined with each other. After setting the initial parameters of the parameters of the array operation unit, the digital image and the rough edge information image are set in pixel units. Appropriate input is performed sequentially from the conversion to the aforementioned repeated information image to the output of each band pixel 値 of the aforementioned repeated information image, and iteratively proceeds until the digital image and the rough edge information image are no longer input. This makes it possible to easily modify the aforementioned parameters by using a general processor. In addition, the array operation unit does not necessarily need to wait closely for the reception of the nearby images of the various images transmitted from the nearby arrangement device. This is because when the aforementioned nearby pixels of various aforementioned images cannot be received from the aforementioned array operation unit located nearby, the aforementioned array operation unit waiting to be received can be replaced with a pixel 相当于 equivalent to 0. At this time, although there may be some noise in the pixels of the various images generated by the array operation unit, in the present invention, almost all the noises are absorbed by the aforementioned various mechanisms. By using a mechanism that replaces the aforementioned band pixels corresponding to 0, edge processing and timeout processing can be implemented simultaneously and simply. In the present invention, an edge information image can be input to form the rough edge information image. In order to generate the aforementioned repeated information image of the aforementioned position and the aforementioned size of the object area indicated by the aforementioned formation of the edge information image, this paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm) 540010 Economy Printed by the Intellectual Property Bureau employee consumer cooperative A7 B7 V. Description of the invention (&) The array operation units are arranged in a grid shape, so that the adjacent array operation units are combined with each other, and the parameters of the array operation units are to be set. After the initial frame, the digital image and the rough-edge information image are appropriately input in pixel units, and the output is sequentially converted from the repeated information image to the pixel information of each frequency band of the repeated information image. Repeat until the digital image and the rough edge information image are no longer input. This makes it possible to easily modify the aforementioned parameters using a general processor. Further, the array operation unit 'does not necessarily need to wait closely for reception of nearby images of the various images transmitted by the nearby arrangement device. This is because when the foregoing nearby pixels of various aforementioned images cannot be received from the aforementioned array computing unit located nearby, the aforementioned array computing unit waiting to be received can be replaced with a pixel equivalent of 0. At this time, although there may be some noise in the pixels of the various images generated by the array operation unit ', in the present invention, almost all noise is absorbed by the various mechanisms described above. By using a mechanism which is replaced by the aforementioned pixel 値 equivalent to 0, edge processing and timeout processing can be realized simultaneously and simply. In addition, the present invention may input the aforementioned image of the object area instead of the aforementioned image of the rough edge information. In order to generate the edge information map, the aforementioned position of the aforementioned object field and the aforementioned repeated animal news of the aforementioned size, the array operation units are arranged in a grid shape, and adjacent array operation units are combined with each other. After setting the initial parameters of the parameters of the array computing unit, the digital image and the rough edge information image are appropriately input in pixel units, and sequentially transformed from the repeated information to the aforementioned repeated information. Figure 27 This paper scale applies China National Standard (CNS) A4 Specification (210 X 297 male f ^^ --- ---, ---, ----------- I--order --------- (Please read the notes on the back before filling this page) Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs and printed on 540010 A7 B7 V. Description of Invention (vL) image to the output of the pixel 値 of each band in the repeated information image, Repeat until the digital image and the rough edge information image are no longer input. According to this, a general processor can be used to easily modify the parameters. In addition, the array operation unit does not necessarily need to wait closely. Attach Reception of nearby images of various aforementioned images transmitted by the aforementioned arranging device. This is because the aforementioned nearby array pixels of various aforementioned images that cannot be received from the aforementioned array computing unit that are located nearby are waiting to receive the aforementioned array computing unit. It is possible to use a relationship equivalent to a pixel 的 of 0. At this time, although there may be some noise in the pixels of each of the foregoing images generated by the foregoing array operation unit, in the present invention, almost all the noise is It is absorbed by the aforementioned various mechanisms. By using a mechanism which is replaced by the aforementioned band pixel unit equivalent to 0, edge processing and timeout processing can be realized simultaneously and simply. The invention according to the tenth aspect of the patent application is a visual device which It is characterized in that each array computing unit arranged in a grid shape in a data processing device of a mechanism capable of detecting the position and size of an object field is provided with a mechanism for initializing the aforementioned array computing unit; if there is no aforementioned digital map that should be input When the image is taken, the processing is ended. Input the image of each band of the rough edge information image. Elementary mechanism; mechanism for inputting band pixels 値 of the object domain image and each band pixel of the digital image; separating the band pixel 値 of the object domain image and the band pixel 値 of the digital image 値In order to generate the updated image of the object area, the paper size is 28. This paper size is applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) '--- * --------------- --- Order --------- line (Please read the precautions on the back before filling this page) 540010
經濟部智慧財產局員工消費合作社印制衣 五、發明說明（Jj) 帶像素値及更新數位圖像之各頻帶像素値的機構； 將自前述更新物體領域圖像所計算之移動量圖像化爲 移動量圖像之頻帶像素値的機構； 隨前述移動量圖像之前述頻帶像素所指之移動位置的 重複數，生成能移動圖像之頻帶像素値的機構； 隨前述能移動圖像的判定，將前述更新物體領域圖像 之前述頻帶像素値移動到前述移動薄； 配合前述更新物體領域圖像之頻帶像素値的移動，移 動前述更新資料圖像之前述各頻帶像素値的機構； 用前述物體區域所包含之附近頻帶像素値之平均値， 以內插法推出前述物體領域所未包含之前述更新物體領域 圖像的前述頻帶像素値的機構； 配合前述更新物體領域圖像之頻帶像素値的內插，以 內插法推出前述更新數位圖像之前述各頻帶像素値的機構 ;以及 輸出內插前述更新圖像後所生成之規格化圖像之各頻 帶像素値的機構。 也就是說，此係爲了將前述陣列運算單元所提供的前 述規格化圖像的生成機能以數位技術來加以實現之算法的 組裝形態。將前述陣列運算單元配置成網格狀，使鄰近之 前述陣列運算單元相互結合，待設定前述陣列運算單元的 各參數的初始値後，將前述物體領域圖像及前述數位圖像 以像素單位適當地加以輸入，依序進行前述物體領域圖像 與前述數位圖像之分離，到前述規格化圖像之各頻帶像素 29 本紙張尺度適用中國國家標準（CNS)A4規格(:210 X 297公釐） —*— --------------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 540010 經濟部智慧財產局員工消費合作社印製 A7 __B7___ 五、發明說明（d ) 値的輸出，反覆進行至前述物體領域圖像及前述數位圖像 不再輸入爲止。據此，能利用一般的處理器，容易地進行 前述參數之修正。又，前述陣列運算單元，未必需要嚴密 進行等待位於附近之前述排列裝置所傳送來的各種前述圖 像之附近圖像的接收。此係因爲無法接收自位於附近之前 述陣列運算單元之各種前述圖像的前述附近像素時，等待 接收之前述陣列運算單元能使用相當於0之像素値加以取 代的關係。此時，雖然前述陣列運算單元所生成之各種前 述圖像的像素中可能有若干的雜訊，但在本發明中，幾乎 所有的雜訊都被前述各種機構加以吸收。藉使用相當於0 之前述頻帶像素値加以取代的機構，能同時且簡單地實現 邊緣處理與超時處理。 申請專利範圍第11項之發明，係一種視覺裝置，其特 徵在於，針對辨識規格化圖像之機構中實現圖案匹配的資 料處理裝置中之配置成網格狀之各個陣列運算單元，係具 備： 將前述陣列運算單元以網格狀加以配置的機構； 將前述陣列運算單元初始化的機構； 至應輸入之模板圖像完全沒有爲止，輸入前述模板圖 像之頻帶像素値的機構； 若無應輸入之前述規格化圖像時，即結束處理的機構 輸入前述規格化圖像之頻帶像素値的機構； 計算匹配結果的機構； 30 —.—---------------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 本紙張尺度適用中國國家標準（CNS)A4規格4210 X 297公釐） 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明（y1) 更新匹配結果圖像的機構；以及 輸出前述匹配結果圖像之頻帶像素數的機構。 也就是說，此係在辨識前述圖像辨識機構中，爲了用 數位技術來實現前述陣列運算單元所提供之圖形匹配之算 法的組裝形態。將前述陣列運算單元配置成網格狀，使鄰 近之前述陣列運算單元相互結合，待設定前述陣列運算單 元的各參數的初始値後，將前述模板圖像及前述規格化圖 像以像素單位適當地加以輸入，依序進行自前述匹配結果 之計算，到前述匹配結果圖像之各頻帶像素値的輸出，反 覆進行至前述規格化圖像不再輸入爲止。據此，能利用一 般的處理器，容易地進行前述參數之修正。又，前述陣列 運算單元，未必需要嚴密進行等待位於附近之前述排列裝 置所傳送來的各種前述圖像之附近圖像的接收。此係因爲 無法接收自位於附近之前述陣列運算單元之各種前述圖像 的前述附近像素時，等待接收之前述陣列運算單元能使用 本身之前述頻帶像素値加以取代的關係。此時，雖然前述 陣列運算單元所生成之各種前述圖像的像素中可能有若干 的雜訊，但在本發明中，幾乎所有的雜訊都被前述各種機 構加以吸收。藉使用本身之前述頻帶像素値加以取代的機 構’能同時且簡單地實現邊緣處理與超時處理。 申請專利範圍第12項之發明，係一種視覺裝置，其特 徵在於，針對使用形成邊緣資訊圖像以分離物體領域機構 的資料處理裝置中之配置成網格狀之各個陣列運算單元， 係具備： 31 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） ---------------------η-------- (請先閱讀背面之注意事項再填寫本頁} 540010 Α7 Β7 五、發明說明（^0) 將前述陣列運算單元以網格狀加以配置的機構； 將前述陣列運算單元中的非線性振盪器與位於前述非 線性振盪器附近之前述非線性振盪器，以結合値加以連接 的機構； 將前述陣列運算單元初始化的機構； 若無應輸入之前述形成邊緣資訊圖像時，即結束處理 的機構； 輸入前述形成邊緣資訊圖像之頻帶像素値的機構； 計算干擾的機構； 計算前述非線性振盪器附近輸入合計的機構； 計算前述非線性振盪器參數的機構； 計算前述非線性振盪器輸出的機構； 計算輪廓參數的機構 計算交界參數的機構；以及 輸出包含有以前述非線性振盪器所分離出之前述物體 領域的物體領域圖像的頻帶像素値的機構。 也就是說，此係爲了以數位技術來實現前述陣列運算 單元所提供之前述物體領域圖像的生成機能之算法的組裝 形態。將前述陣列運算單元配置成網格狀，使鄰近之前述 陣列運算單元相互結合，待設定前述陣列運算單元的各參 數的初始値後，將前述形成邊緣資訊圖像以像素單位適當 地加以輸入，依序進行自前述干擾之計算，到前述物體領 .域圖像之各頻帶像素値的輸出，反覆進行至前述形成邊緣 資訊圖像不再輸入爲止。據此，能利用一般的處理器，容 32 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐) (請先閱讀背面之注意事項再填寫本頁) |裝----Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. 5. Description of the Invention (Jj) A mechanism with a pixel 値 and a pixel 各 in each frequency band for updating the digital image; a graph of the amount of movement calculated from the previously updated object area image Is the mechanism of the band pixel 値 of the moving image; the mechanism of generating the band pixel 値 of the moving image according to the repetition number of the moving position of the band pixel of the moving image; It is determined that the band pixel 前述 of the update object area image is moved to the mobile thin film; a mechanism for moving the band pixel 更新 of the update data image in accordance with the movement of the band pixel 値 of the update object area image; The mean 値 of the nearby band pixels 包含 included in the aforementioned object region is derived by interpolation from the aforementioned means of updating the aforementioned band pixels 値 of the object domain image which are not included in the aforementioned object domain; Interpolation, the interpolation mechanism is used to introduce the aforementioned mechanism for updating the aforementioned pixel 値 of each frequency band of the digital image; and Each frequency band generated by the interpolation after the updated image of the normalized image pixel Zhi mechanism. In other words, this is an assembly form of an algorithm for realizing the aforementioned normalized image generation function provided by the aforementioned array operation unit with digital technology. The array operation unit is arranged in a grid shape, so that adjacent array operation units are combined with each other. After setting the initial parameters of the parameters of the array operation unit, the object area image and the digital image are appropriately set in pixel units. Input them in sequence, and sequentially separate the aforementioned image of the object area from the aforementioned digital image to the pixels of each frequency band of the aforementioned normalized image. 29 This paper size applies the Chinese National Standard (CNS) A4 specification (: 210 X 297 mm) ) — * — -------------- Order --------- line (please read the precautions on the back before filling this page) 540010 Employee Consumption of Intellectual Property Bureau, Ministry of Economic Affairs The cooperative prints A7 __B7___ 5. The output of invention description (d) 値 is repeated until the aforementioned image of the object area and the aforementioned digital image are no longer input. This makes it possible to easily modify the aforementioned parameters by using a general processor. In addition, the array operation unit does not necessarily need to strictly wait for the reception of nearby images of the various kinds of images transmitted by the nearby arrangement devices. This is because when it is impossible to receive the aforementioned nearby pixels of various aforementioned images located near the aforementioned array operation unit, the aforementioned array operation unit waiting to be received can use a pixel equivalent of 0 to replace it. At this time, although there may be some noise in the pixels of the aforementioned images generated by the aforementioned array operation unit, in the present invention, almost all of the noise is absorbed by the aforementioned various mechanisms. By using a mechanism in which the aforementioned band pixels 値 corresponding to 0 are replaced, edge processing and timeout processing can be realized simultaneously and simply. The invention of claim 11 in the scope of patent application is a visual device characterized in that, for each array computing unit arranged in a grid shape in a data processing device that realizes pattern matching in a mechanism for identifying a normalized image, it has: A mechanism for arranging the aforementioned array operation unit in a grid shape; a mechanism for initializing the aforementioned array operation unit; a mechanism for inputting the band pixel 値 of the aforementioned template image until there is no template image to be input; if there is no input In the case of the aforementioned normalized image, the mechanism that ends processing inputs the mechanism of the band pixel 値 of the aforementioned normalized image; the mechanism that calculates the matching result; 30 —.—-------------- -Order --------- Line (Please read the precautions on the back before filling this page) This paper size applies to China National Standard (CNS) A4 specification 4210 X 297 mm) Consumption by employees of the Intellectual Property Bureau of the Ministry of Economic Affairs Cooperatives printed 540010 A7 B7 5. Description of the invention (y1) A mechanism for updating the matching result image; and a mechanism for outputting the number of pixels in the frequency band of the aforementioned matching result image. In other words, this is an assembly form for recognizing the aforementioned image recognition mechanism in order to use digital technology to realize the pattern matching algorithm provided by the aforementioned array operation unit. The array operation unit is arranged in a grid shape, so that the adjacent array operation units are combined with each other. After setting the initial parameters of the parameters of the array operation unit, the template image and the normalized image are appropriately set in pixel units. The input is performed sequentially, and the calculation from the foregoing matching result is sequentially performed to the output of pixels 値 in each frequency band of the foregoing matching result image, which is repeatedly performed until the normalized image is no longer input. This makes it possible to easily perform the correction of the aforementioned parameters using a general processor. In addition, the array operation unit does not necessarily have to wait closely for reception of nearby images of various images transmitted by the nearby arrangement device. This is because when the aforementioned nearby pixels of various aforementioned images cannot be received from the aforementioned array operation unit located nearby, the aforementioned array operation unit waiting to be received can be replaced by the aforementioned band pixel 频带 of itself. At this time, although there may be some noise in the pixels of the various images generated by the array operation unit, in the present invention, almost all the noises are absorbed by the various mechanisms described above. The mechanism ', which is replaced by the aforementioned band pixels 本身, can simultaneously and simply implement edge processing and timeout processing. The invention of claim 12 in the scope of patent application is a visual device, which is characterized in that, for each array operation unit arranged in a grid shape in a data processing device that uses an edge information image to separate an object field mechanism, the device includes: 31 This paper size applies to China National Standard (CNS) A4 (210 X 297 mm) --------------------- η -------- (Please read the precautions on the back before filling out this page} 540010 Α7 Β7 V. Description of the Invention (^ 0) The mechanism for arranging the aforementioned array operation unit in a grid shape; the non-linear oscillator in the aforementioned array operation unit and A mechanism for connecting the aforementioned non-linear oscillator near the aforementioned non-linear oscillator in conjunction with 値; a mechanism for initializing the aforementioned array operation unit; a mechanism for ending processing if there is no aforementioned input of an edge information image; A mechanism for inputting the aforementioned pixel bands forming a band information image; a mechanism for calculating interference; a mechanism for calculating a total input near the nonlinear oscillator; a mechanism for calculating the parameters of the nonlinear oscillator; A mechanism for outputting a nonlinear oscillator; a mechanism for calculating a contour parameter; a mechanism for calculating a boundary parameter; and a mechanism for outputting a band pixel 値 that includes an image of the object area of the object area separated by the nonlinear oscillator. That is, this is an assembly form of an algorithm for realizing the function of generating the image of the object area provided by the array operation unit with digital technology. The array operation unit is arranged in a grid shape so that the array operation unit adjacent to the array operation unit is arranged. Combined with each other, after setting the initial parameters of the parameters of the aforementioned array operation unit, the aforementioned formed edge information image is appropriately input in pixel units, and the calculation from the aforementioned interference is sequentially performed to the aforementioned object domain image. The output of pixels in each frequency band is repeated until the aforementioned formation of the edge information image is no longer input. Based on this, a general processor can be used to accommodate 32 paper sizes that comply with the Chinese National Standard (CNS) A4 specification (210 X 297 cm). Li) (Please read the notes on the back before filling this page) |
訂---------線I 經濟部智慧財產局員工消費合作社印製 經濟部智慧財產局員工消費合作社印製 540010 Α7 Β7 五、發明說明（3ί) 易地進行前述參數之修正。又，前述陣列運算單元，未必 需要嚴密進行等待位於附近之前述排列裝置所傳送來的各 種前述圖像之附近圖像的接收。此係因爲無法接收自位於 附近之前述陣列運算單元之各種前述圖像的前述附近像素 時，等待接收之前述陣列運算單元能使用本身之參數値或 像素値加以取代的關係。此時，雖然前述陣列運算單元所 生成之各種前述圖像的像素中可能有若干的雜訊，但在本 發明中，幾乎所有的雜訊都被前述各種機構加以吸收。藉 使用本身之前述參數値或前述頻帶像素値加以取代的機構 ，能同時且簡單地實現邊緣處理與超時處理。 申請專利範圍第13項之發明，係一種視覺裝置，其特 徵在於，針對具有：資料輸入機構、依序記憶前述資料的 機構、於陣列運算單元間傳送前述資料的機構、使用前述 資料進行計算的機構、以及輸出前述資料的機構之陣列運 算單元，係具備： 將前述陣列運算單元配置成網格狀的機構； 根據前述陣列運算單元之各自的位置關係，使附近陣 列運算單元相互結合的機構； 於相鄰之前述陣列運算單元間，進行資料通信的機構 ;以及 使前述各陣列運算單元獨立動作的機構。 針對輸入前述資料的前述機構，依序記憶前述資料的 前述機構，於前述陣列運算單元間傳送前述資料的前述機 構，使用前述資料進行計算的前述機構，以及輸出前述資 33 本紙張尺度適用中國國家標準（CNS)A4規格4210 X 297公釐） —.—,--------------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 540010 A7 _____B7 _ 五、發明說明（Ρ) (請先閱讀背面之注意事項再填寫本頁) 料的前述機構，前述各個陣列運算單元與其網格狀排列之 位置無關的，具有相同的動作步驟。因此’當使用硬體進 行前述陣列運算單元之組裝時’能將實現前述陣列運算單 元的相同電路，以正確之規則配置於平面上’由於該等電 路僅需連接相鄰者即可，故配線量亦較少’只要配合處理 之圖像的尺寸增減電路之數量即可’且能使各個電路並列 動作。又，當利用軟體進行前述陣列運算單元之組裝時， 可使用並列性高的程式，執行將前述陣列運算單元網格狀 並列的視覺裝置。 申請專利範圍第14項之發明，係一種視覺裝置，其特 徵在於： 陣列運算單元係具備：具有輸入資料處理機構的處理 器，用來記憶前述資料之處理程式與變數的記憶體，以及 用來與相鄰之陣列運算單元進行通信的控制器； 前述控制器係具備：將輸入之前述資料記憶在前述記 憶體中的機構，將前述記憶體中之前述變數傳送到前述陣 列運算單元的機構，以及將相鄰之前述陣列運算單元所傳 送的前述變數記憶在前述記憶體的機_ ° 經濟部智慧財產局員工消費合作社印製 也就是說，係以硬體組裝前述陣列運算單元的電路。 前述陣列運算單元，可使用用來處理輸入之前述資料的一 般處理器，與用來記憶處理前述資料之程式與變數的一般 記憶體。前述控制器，於前述陣列運算單元爲與最大4附 近者相互結合時，只要對相鄰之前述陣列運算單元傳送變 數即可，但當前述陣列運算單元必須與8附近以上者相互 34 本紙張尺度適用中國國家標準（CNS)A4規格（,210 X 297公釐） 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明） 結合時，由於不包含在4附近者之前述陣列運算單元的前 述變數，暫時傳送至相鄰之前述陣列運算單元，因此將前 述前述變數再傳送給自己，即就能加以接收。又，利用將 前述記憶體中之前述各變數傳送至相鄰之前述陣列運算單 元的前述機構，亦能將自己的前述變數傳送至到未包含於 4附近者的前述陣列運算單元。據此，前述陣列運算單元 ，就硬體裝置而言，雖僅與相鄰之前述陣列運算單元接線 ，亦能與8附近以上之前述陣列運算單元適當的進行前述 資料之通信。又，能適當地解決前述課題中，有關硬體組 裝及即實處理的各種問題。 [圖式之簡單說明] 第1圖係自框圖像計算移動物體數量之視覺裝置的方 塊圖。 第2圖係使用物體/背景分離機構，自框圖像計算移動 物體數量之視覺裝置的方塊圖。 第3圖係自框圖像計算所有物體數量之視覺裝置的方 塊圖。 第4圖係使用物體/背景分離機構，自框圖像計算所有 物體數量之視覺裝置的方塊圖。 第5圖係用來求出移動物體與靜止物體比例之視覺裝 置的方塊圖。 第6圖係用來控制移動式攝像機之視覺裝置的方塊圖 〇 第7圖係生成物體規格化圖像之視覺裝置的方塊圖。 35 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） ----------------------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 540010 A7 B7 五、發明說明 第8圖係輸出物體辨識結果之視覺裝置的方塊圖。 弟9圖係生成ΪΕ境地圖之視覺裝置的方塊圖。 第10圖係使用環境地圖，控制移動式攝像機之視覺裝 置的方塊圖。 第11圖係計算任意物體數量之視覺裝置的方塊圖。 第12圖係利用幾何解析，進行高速化之視覺裝置的方 塊圖。 第13圖係網格狀配置之陣列運算單元的方塊圖。 第14圖係顯示本實施形態之圖像記憶機構之算法的流 程圖。 第15圖係顯示本實施形態之圖像振動機構之算法的流 程圖。 第16圖係顯示本實施形態之邊緣資訊生成機構之算法 的流程圖。 第17圖係使用數位圖像，將粗邊緣資訊形成爲形成邊 緣資訊時之說明圖。 第18圖係顯示本實施形態之邊緣資訊形成機構之算法 的流程圖。 第19圖係將自低解像度數位圖像所生成之低解像度粗 邊緣資訊形成爲形成邊緣資訊時之說明圖。 第20圖係將自低解像度數位圖像所生成的低解像度粗 邊緣資訊領域切出後，形成爲形成邊緣資訊時的說明圖。 第21圖係檢測邊緣資訊圖像中之物體位置及大小時的 說明圖。 36 (請先閱讀背面之注意事項再填寫本頁) |裝--------訂---------線 經濟部智慧財產局員工消費合作社印製 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明 第22圖係顯示本實施形態之位置/大小檢測機構之算 法的流程圖。 第23圖係檢測物體領域圖像中之物體位置及大小時的 說明圖。 第24圖係將數位圖像之分離物體領域加以規格化時的 說明圖。 第25圖係顯示本實施形態之領域規格化機構之算法的 流程圖。 第26圖係顯示本實施形態之規格化圖像保持機構之算 法的流程圖。 第27圖係相對規格化圖像，自模板圖像中進行圖形匹 配時的說明圖。 第28圖係顯示本實施形態之圖像辨識機構中圖形匹配 ‘ 之算法的流程圖。 第29圖係顯示三角形之邊緣資訊分離爲三角形內側領 域與外測領域之狀態的說明圖。 第30圖係顯示本實施形態之物體/背景分離機構之算 法的流程圖。 第31圖係顯示虛線狀態之三角形之邊緣資訊分離爲虛 線三角形內側領域與外測領域之狀態的說明圖。 第32圖係顯示重疊兩個三角形之邊緣資訊分離爲兩個 三角形領域與背景領域之狀態的說明圖。 第33圖係顯示重疊兩個圓形虛線狀態之邊緣資訊分離 爲兩個圓形領域與背景領域之狀態的說明圖。 37 玉紙張尺度適用中國國家標準（CNS)A4規格4210 X 297公釐1 一 -------------ΦΜί----- 丨訂---------線 (請先閱讀背面之注意事項再填寫本頁) 540010 經濟部智慧財產局員工消費合作社印製 A7 Β7 五、發明說明（#) 第34圖係陣列運算單元之內部構造的方塊圖。 第35圖係控制器的方塊圖。 第36圖係顯示標記解碼器之輸入輸出信號的說明圖。 第37圖係顯示標記編碼器之輸入輸出信號的說明圖。 第38圖係顯示處理器透過控制器，將資料傳送至相鄰 之的陣列運算單元之算法的流程圖。 第39圖係顯示控制器自相鄰之陣列運算單元接收資料 之算法的流程圖。 第40圖係顯示處理器自上輸入暫存器接收資料之算法 的流程圖。 [發明之最佳實施形態] 以下，對顯示於第1〜12圖之12個視覺裝置的實施形 態加以說明。該等實施形態係利用圖像取得機構11(參照第 1圖）、圖像記憶機構12(參照第1圖）、圖像振動機構13(參 照第3圖）、邊緣資訊生成機構14(參照第1圖及第3圖）、 邊緣資訊形成機構15(參照第1圖）、物體/背景分離機構 16(參照第2圖）、位置/大小檢測機構17 (參照第1圖及第2 圖）、像素計數機構18(參照第1圖）、像素保持機構丨9(參 照弟1圖及第5圖)等。圖像取得機構11，是接收攝像機白勺 攝像信號(框圖像1)，將之轉換爲適當格式與大小的數位圖 像111 ;圖像記憶機構12是將數位圖像丨11記憶一定期間 ;圖像振動機構13是使用數位電路，ί吏數位圖像in振重力 ;邊緣資訊生成機構14是根據兩種數位圖像in，生成移 動物體2或靜止物體3的粗邊緣資訊112 ;邊緣資訊形成 38 ----------------------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 本紙張尺度適用中國國家標準（CNS)A4規格4210 x 297公釐） —— 經濟部智慧財產局員工消費合作社印製 540010 Α7 ----- Β7 五、發明說明（q) 機構15是將粗邊緣資訊U2形成爲更明確的形成邊緣資訊 114 ;物體/背景分離機構16是將形成邊緣資訊U4所區分 的區域加以分離；位置/大小檢測機構17是檢測形成邊緣 資訊114所區分或分離出之各區域的位置及大小；像素計 數機構18是自具有適當大小區域的位置，計算出區域數量 ;像素保持機構19是輸出區域數量或區域數量的比例。 如第1圖所示，移動物體計數部101將攝像機所拍攝 之動態，圖像的框圖像1，輸入圖像取得機構11，依照圖像 記憶機構12、邊緣資訊生成機構14、邊緣資訊形成機構 15、位置/大小檢測機構17、像素計數機構18、像素保持 機構19的順序進行輸出入，輸出映在框圖像1之移動物體 2的數量。又，框圖像1亦可是用數位相機連續拍攝的靜 止圖像。 圖像取得機構11從攝像機輸入動態圖像的框圖像1時 ，若動態圖像爲類比信號時，係使用一般的畫像框獲取板 對框圖像1進行A/D轉換以轉換爲數位信號，形成數位圖 像111。此外，特別是若能直接輸入CCD攝像元件等之電 壓，亦可利用A/D轉換轉換爲適當位元數的數位信號。當 動態圖像爲數位信號時，若有被壓縮時，即加以展開，若 未被壓縮，即原封不動地加以輸入。利用這種方法，因能 切出動態圖像中任意的框圖像1，故切出該框圖像1形成 數位圖像111。由於轉換後之數位圖像111會根據適當的 格式而具有任意的圖像尺寸，故圖像取得機構Η能用像素 單位，將圖像資料轉換爲能參照的格式，於移動物體計數 39 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） --1!-----------------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 540010 A7 B7 經濟部智慧財產局員工消費合作社印製 五、發明說明（4) 部101切出所需之圖像尺寸，作爲數位圖像111加以輸出 。若圖像取得機構11能並行輸出數位圖像111的所有像素 的話，自圖像取得機構11至圖像記憶機構12之通信，即 能以各像素並行進行。 當圖像記憶機構12從圖像取得機構11輸入數位圖像 111時，即配合移動物體計數部101的時間分辨能力或各 機構的計算能力，將數位圖像111記憶一定時間。也就是 說，由於在此一定時間中，即使輸入數位圖像111，圖像 記憶機構12也不變更記憶圖像，因此後續各機構能在不同 的時點輸入相同的數位圖像111。而且，圖像記憶機構12 對數位圖像111並不施以圖像處理，故對數位圖像111的 所有像素，係在保持二次元之相位關係下加以記憶。若圖 像記憶機構12能並行輸出數位圖像111之所有像素，自圖 像記憶機構12至邊緣資訊生成機構14之通信，即能以各 像素並行進行。 當邊緣資訊生成機構14從圖像記憶機構12輸入數位 圖像111時，藉比較先前所輸入之數位圖像111，生成移 動物體2之粗邊緣資訊圖像113。由於邊緣資訊生成機構 14能僅在各像素的附近處理時生成粗邊緣資訊圖像113， 因此適合並行化。若邊緣資訊生成機構14能並行輸出粗邊 緣資訊圖像113之所有像素，自邊緣資訊生成機構14至邊 緣資訊形成機構15之通信，即能各像素並列進行。 當邊緣資訊形成機構15自邊緣資訊生成機構14輸入 粗邊緣資訊圖像113，即參照圖像記憶機構12所記憶的數 40 ---T . (請先閱讀背面之注意事項再填寫本頁) 言 Γ 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 540010 A7 B7 五、發明說明（4) (請先閱讀背面之注意事項再填寫本頁) 位圖像111，生成比數位圖像113更明確且明瞭的移動物 體2之形成邊緣資訊圖像115。由於邊緣資訊形成機構15 能僅在各像素的附近處理生成形成邊緣資訊圖像丨15’因 此適合並列化。若邊緣資訊形成機構15能並列輸出形成邊 緣資訊圖像115的所有像素，自邊緣資訊形成機構15至位 置/大小檢知機構17之通信，即能各像素並列進行。 當位置/大小檢測機構Π自邊緣資訊形成機構15輸入 形成邊緣資訊圖像115時，即檢測形成邊緣資訊114所指 示之移動物體2的領域位置及大小。由於位置/大小檢知機 構17僅用各像素附近處理，即能生成顯示移動物體2之領 域位置及大小之檢測結果的重複資訊圖像132,因此適合 並列化。若位置/大小檢知機構Π能並列輸出重複資訊圖 像132的所有像素，自位置/大小檢知機構Π至像素計數 機構18之通信，即能各像素並列進行。 經濟部智慧財產局員工消費合作社印製 當像素計數機構18自位置/大小檢知機構17輸入重複 資訊圖像132時，即能計算顯示適當大小之移動物體2之 領域位置的像素數量。該處理，與其使用並列處理，不如 使用逐次處理來進行較爲方便。因此，重複資訊圖像132 的各像素係記憶於線性記憶體中。顯示移動物體2之領域 位置的像素數量，係自像素計數機構18輸出至像素數保持 機構19。 當像素數量保持機構19自像素計數機構18輸入像素 數量時，即將像素數量轉換爲2的補數表現或浮動小數點 表現等輸出端所需的形式，進行必要期間之輸出。也就是 41 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 540010 經濟部智慧財產局員工消費合作社印製 A7 B7___ 五、發明說明（蛉。） 說，移動物體計數部101能輸出移動物體數量。 前述各機構，若能用獨立的硬體組裝的話，即能實施 並行計算，故移動物體計數部101能具有即時性。因此， 適用於計算移動速度快的物體數時，或以短時間處理大量 的動態圖像時。另外，若幾個機構能利用軟體組裝於—個I 處理器上的話，則計算速度雖然變慢但卻能價廉地加以製 造。因此，非常適合即使輸出計數結果要花費數分鐘以上 也無所謂的應用例。 當位置/大小檢測機構17使用邊緣資訊形成機構15所 生成的形成邊緣資訊圖像115，檢測移動物體2的領域位 置及大小時，有可能因移動物體2之密度，使顯示移動物 體2之領域位置的像素數量與移動物體2之數量不同。主 要原因是，形成邊緣資訊114不一定是正確地抓出移動物 體2之邊緣，以及位置/大小檢測機構17無法從形成邊緣 資訊114辨別移動物體2的形狀。因此，若移動物體2的 密度較高，就會組合不同移動物體2所生成的形成邊緣資 訊114，有時會與不存在的物體邊緣資訊混在一起。爲了 解決這種問題’雖能根據形成邊緣資訊114來辨別移動物 體2之形狀，但以往的幾何解析方法由於係進行大區域處 理，因此計算量變多，若欲提高辨別結果之精密度，計算 時間即就會以指數函數變長。因此，利用附近處理，可使 用物體/背景分離機構16(參照第2圖)作爲解決該問題的機 構。 當物體/背景分離機構16自邊緣資訊形成機構15輸入 42 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） —1—*--------------訂---------線 (請先閲讀背面之注意事項再填寫本頁) 經濟部智慧財產局員工消費合作社印製 540010 A7 -------B7 五、發明說明（ί^Ι) 形成邊緣資訊圖像115時，即能將物體領域141所包含的 像素與背景所包含的像素，分離成不同的組，將分離結果 以組單位依序輸出。又，當物體領域141即使相鄰而能以 形成邊緣資訊114明確加以區別時，物體/背景分離機構16 ’能將該等物體領域分離成不同的組。因此，組數也達到 3以上。由於物體/背景分離機構16能僅以附近處理分離出 各像素中之物體領域141與背景，因此適合並列化。若物 體/背景分離機構16能並列輸出物體領域圖像142的所有 像素’自物體/背景分離機構16至位置/大小檢測機構17之 通信，即能各像素並列進行。 因此’如第2圖所示，藉使用物體/背景分離機構16， 即能僅以附近處理一面補足形成邊緣資訊114, 一面分離 移動物體2的領域與其他的背景領域。此時，若移動物體 2的領域類似圓形的話，換句話說，若移動物體2的領域 中凹凸較少的話，位置/大小檢測機構17就能抓出一定大 小的領域，以特定位置。因此，若能事前知道移動物體2 係類似圓形的話，像素計數機構18即能藉使用物體/背景 分離機構16，能計算顯示特定大小之移動物體2之領域位 置的像素數。也就是說，移動物體計數部1〇1能輸出精度 更高的移動物體數量。 以上，說明了計算動態圖像之二個以上的框圖像1中 的移動物體2之數量時，移動物體計數部1〇1是如何動作 。接著，說明動態圖像之一個框圖像1中的移動物體2與 靜止物體3的總數，也就是說，計算所有物體數量時之情 43 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） --------,--------------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 540010 A7 ------- B7 五、發明說明（私） 形。 / 首先’基本動作與使用前述動態圖像之框圖像1計算 移動物體2之數量的情形相同。也就是說，用適當的方法 將框圖像1中之靜止物體3看起來是替換爲框圖像1中的 移動物體2即可。如此，由於會生成靜止物體3之粗邊緣 資訊112，因此移動物體計數部1〇1亦能計算靜止物體3的 數量。將靜止物體3視爲移動物體2之最簡單的方法，即 係使用振動台等使攝像機(數位相機)或靜止物體3本身進 行微細振動。但是，因該方法必須要有物理性機構，故移 動物體計數部101將會變的昂貴或複雜。另一方面，作爲 不使用物理性機構的機構，有圖像振動機構13(參照第3圖 )° 當圖像振動機構13自圖像記憶機構12輸入數位圖像 111時，靜止物體3即在數位圖像111中，以圖像單位一起 或像素單位個別之方式，在3像素前後的範圍上下左右地 振動。若圖像振動機構13能並列輸出數位圖像111的所有 像素，自圖像振動機構13至邊緣資訊生成機構14之通信 ，即能各像素並列進行。 因此，如第3圖所示，圖像振動機構13，即使圖像取 得機構11所生成之數位圖像111中的靜止物體3，在3像 素前後範圍上下左右地振動。據此，由於邊緣資訊生成機 構14，就能將靜止物體3視爲移動物體2，以生成靜止物 體3的粗邊緣資訊112，故所有物體計數部102，即能計算 移動物體2與靜止物體3的總數，也就是說，能計算所有 44 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） —.—·------ΛΜ, (請先閱讀背面之注意事項再填寫本頁) 訂---------線 « 經濟部智慧財產局員工消費合作社印製 540010 經濟部智慧財產局員工消費合作社印製 A7 五、發明說明（於)） 物體數量。 再者，與移動物體2及靜止物體3無關地，物體/背景 分離機構16僅利用形成邊緣資訊U4分離物體領域141與 背景領域。因此，若能預先知道移動物體2及靜止物體3 係類似於圓形的話，則如第4圖所示’將邊緣資訊形成機 構15所生成的形成邊緣資訊圖像丨15輸入物體/背景分離 機構16，將物體/背景分離機構16所生成的物體領域圖像 142輸入位置/大小檢測機構17 ’據此’位置/大小檢測機構 17即能計算顯示特定大小的移動物體2及靜止物體3之領 域位置的像素數量。也就是說’所有物體計數部丨02能輸 出更高精度的所有物體數量。 另外，如上所述，視覺裝置使用計算動態圖像的框圖 像1中的移動物體2之數量的移動物體計數部101，與大 致具備相同機構的所有物體計數部102,即能計算動態圖 像1中的移動物體2及靜止物體3的總數。因此，考慮使 用一個視覺裝置，同時計算移動物體2及靜止物體3的方 法。此時，若能求出移動物體2及靜止物體3之數量的話 ，即能對所有物體容易地計算出移動物體2的比例，或靜 止物體3的比例。以下，亦對用以求出移動物體2及靜止 物體3之比例的視覺裝置加以說明。 求出移動物體2與所有物體的數量，且求出移動物體 2及靜止物體3之比例之視覺裝置的整體構成，如第5圖 所示’大致劃分爲移動物體計數部1〇1與所有物體計數部 102所組成。移動物體計數部101與所有物體計數部ι〇2， 45 ----------------------訂—-------—線 (請先閱讀背面之注意事項再填寫本頁) 本紙張尺度適用中國國豕標準（CNS)A4規格（21〇 X 297公餐） 540010 A7 B7____ 五、發明說明（¥() (請先閱讀背面之注意事項再填寫本頁) 是由前述各個機構所構成。又’移動物體計數部101與所 有物體計數部102的圖像取得機構11及圖像記憶機構12， 由於係進行完全相同的動作’故在第5圖中，係共用圖像 取得機構11及圖像記億機構12 °因此’重要的是移動物 體計數部101的像素數保持機構19 ’是從所有物體計數部 102的像素計數機構18輸入所有物體數’所有物體計數部 102的像素數保持機構19，是從移動物體計數部101的像 素計數機構18輸入移動物體數’利用比例切換信號等來自 外部的信號，計算移動物體2及靜止物體3的比例進行切 換以變換爲浮動小數點表記等。當然’若不需計算移動物 體2及靜止物體3的比例’或能利用別的裝置來實現的話 ，就不必將此機能加於像素數保持機構19。相反的，當同 時求移動物體數及靜止物體數時’只要在像素數保持機構 19追加對應之機能即可。藉在像素數保持機構19中使用 一般處理器，視覺裝置即能視用途自由地進行此種變更。 經濟部智慧財產局員工消費合作社印製 又’如上述之說明般’從申旨靑專利軔圍弟1項到申請 專利範圍第3項之視覺裝置(由移動物體計數部101及所有 物體計數部102.所構成），是用來計算攝像機所拍攝之框圖 像1中之移動物體2及靜止物體3之數量的裝置。因此’ 若移動物體2及靜止物體3是包含在框圖像1中的話，該 等視覺裝置就能計算移動物體2及靜止物體3的數。但是 ，移動物體2及靜止物體3 —旦偏離框圖像1的話，該等 視覺裝置就無法計算移動物體2及靜止物體3的數量。因 此，以下，就使用移動式攝像機1〇(參照第6圖），一面常 46 ^紙張尺度適用中國國家標準（CNS)A4規格¢210 X 297公釐） 經濟部智慧財產局員工消費合作社印製 540010 Α7 Β7 五、發明說明（t〇 時搜索移動物體2及靜止物體3，一面計算其數量之申請 專利範圍第4項到申請專利範圍第5項的視覺裝置加以說 明。 首先，移動式攝像機10備有向水平方向及垂直方向移 動的機構，藉輸入來自外部之控制移動角度的指令，能進 行各個搖動鏡頭及傾斜。又，移動式攝像機10備有變更拍 攝圖像倍率的機構，藉輸入來自外部之控制倍率的指令， 能進行變焦。因此，移動式攝像機10，能根據來自外部的 指令，使攝像機進行微細的振動。據此，移動式攝像機10 所拍攝之動態圖像的框圖像1產生晃動，框圖像1中的物 體被拍攝得宛如振動。也就是說，若使用移動式攝像機10 的話，即不需使用所有物體計數部102的圖像振動機構13 ，僅用移動物體計數部101即能計算所有物體的領域。這 種方法爲了使用移動式攝像機10之物理性機構，在處理速 度與計數精度等方面雖有問題，但由於僅用移動物體計數 部101即能計算移動物體2與所有物體兩者的領域，故適 合計算位於無法包含於框圖像1中之寬敞地方的移動物體 2及靜止物體3。 此處’移動式攝像機10係能將根據搖動鏡頭、傾斜及 變焦等移動指令之所移動之現在位置，視需要加以輸出， 且亦能視需要將現在是移動中或是停止中等移動式攝像機 10的狀態加以輸出者。此時，若視覺裝置能控制移動式攝 像機10的搖動鏡頭、傾斜及變焦的話’該視覺裝置即能變 更移動式攝像機10的方向及倍率’以使移動式攝像機1〇 47 本紙張尺度適用中國國家標準（CNS)A4規格v(210 X 297公爱） ------·--------------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 540010 Α7 Β7 五、發明說明 能常時用適當的大小拍攝移動物體2及靜止物體3。 因此，第6圖係顯不移動式攝像機1〇具備有能常時用 適當的大小來拍攝移動物體2及靜止物體3的基本機構之 申請專利範圍第4項之視覺裝置。此時·，視覺裝置係根據 機能，使用三種座標系統，亦即使用攝像機座標系統、圖 像座標系統及環境座標系統。第一，攝像機座標系統，如 其字義，係移動式攝像機爲了用各個最小控制單位控制移 動鏡頭、傾斜及變焦所使用之攝像機內部的三次元球座標 系統。一般說來，攝像機座標系統的原點，係稱爲原點 (home position)之移動式攝像機的固有位置。攝像機座標系 統是能顯示移動物體2及靜止物體3之物理位置的唯一座 標系統。即使如此，攝像機座標系統，由於在每種移動式 攝像機10中皆不同，故無一般性。因此，視覺裝置須對其 他的機構隱匿攝像機座標系統。第二，圖像座標系統係將 移動式攝像機10所拍攝之框圖像1的中央作爲原點、以像 素爲單位的二次元座標系統。此係用來顯示移動物體2及 靜止物體3究竟是位於框圖像1中的那一像素上。因此， 圖像座標系統，雖適合區別位於框圖像1中複數物體之微 細的位置，但僅藉圖像座標系統是無法顯示移動物體2及 靜止物體3的物理位置。第三，環境座標系統，是視覺裝 置用來在內部邏輯上統一顯示移動物體2及靜止物體3之 位置所使用的三次元球座標系統。環境座標系統在水平方 向及垂直方向使用以弧度(radian)爲單位的角度，爲了顯示 物體大小與離物體距離的積，在距離方向係使用以1.0爲 48 本紙張尺度適用中國國家標準（CNS)A4規格ί210 X 297公釐） (請先閱讀背面之注意事項再填寫本頁) >衣--------訂---------線』 經濟部智慧財產局員工消費合作社印製 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（ep 單位的實數。一般說來，由於物體的大小不會極端地變化 ，故可將到物體之距離與移動式攝像機10的倍率視爲正比 。環境座標系統的原點爲任意。亦即，原則上，環境座標 系統是用來顯示環境座標系統上之任意2點的相對座標。 視覺裝置利用移動式攝像機10,藉將可攝影環境中的物體 投影於環境座標系統中，即能區別複數的物體。 因此，攝像機座標系統及圖像座標系統必須與各個環 境座標系統相互地進行轉換。擔負此角色的機構即係攝像 機/環境座標轉換機構20、圖像/環境座標轉換機構21及運 動控制機構23。該等機構係自移動式攝像機10及圖像取 得機構11的規格，求出攝像機座標系統及圖像座標系統的 各單位，以計算用以轉換爲環境座標系統之矩陣。又，亦 可藉計算自攝像機座標系統到環境座標系統之轉換矩陣的 逆矩陣，求出自環境座標系統到攝像機座標系統的轉換矩 陣。不過，由於攝像機座標系統的原點是移動式攝像機10 的原點位置，因此自攝像機座標系統轉換而來之環境座標 系統之位置，係環境座標系統上移動式攝像機10之原點位 置的相對位置。另一方面，自環境座標系統轉換而來之攝 像機座標系統的位置，係攝像機座標系統上自移動式攝像 機10現在位置的相對位置。再者，由於圖像座標系統是二 次元座標系統，故僅用圖像座標系統是無法轉換爲環境座 標系統。因此，圖像/環境座標轉換機構21，是使用環境座 標系統所顯示之移動式攝像機10的方向及倍率，與框圖像 1中移動物體2及靜止物體3的領域大小，藉隨時計算隨 49 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐) * —---------------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 540010 五 濟 部 智 慧 財 產 局 員 工 消 費 合 作 社 印 製 A7 B7 發明說明（以/) 時轉換矩陣，以自圖像座標系統轉換爲環境座標系統。此 外，自圖像座標系統轉換而來之環境座標系統的位置，成 爲自圖像中心的相對位置。 因此，申請專利範圍第4項之視覺裝置的其他機構， 可視爲爲了常時以適當之大小拍攝移動物體2及靜止物體 3，用來生成及轉換該三個座標系統所顯示之移動物體2及 靜止物體3位置的機構。因此，以座標系統攝像機座標系 統、圖像座標系統及環境座標系統爲基礎，說明第6圖。 首先，圖像取得機構11、邊緣資訊生成機構14及位 置/大小檢測機構14及大小檢測機構π，係與移動物體計 數部101及所有物體計數部102所構成之申請專利範圍第i 項到第3項之視覺裝置中所說明者相同。不過，申請專利 範圍第1項到第3項之視覺裝置中，係將生成邊緣資訊形 成機構15所生成之形成邊緣資訊圖像U5輸入位置/大小 檢測機構17，至於申請專利範圍第4項之視覺裝置中，則 係將生成邊緣資訊生成機構14所生成之粗邊緣資訊_像 113輸入位置/大小檢測機構17。當然，在這種視覺裝麾中 ，雖能藉使用邊緣資訊形成機構丨5，將形成邊緣資訊#像 115輸入fii置/大小檢測機構17，但因下述之理由，該視覺 裝置之性能，即使不使用邊緣資訊形成機構15亦已足夠息 第一，該視覺裝置中，不需要如申請專利範圍第丨項到第 3項之視覺裝置般，計算移動物體2及靜止物體3的領域 數量。反而該視覺裝置中較重要的是，搜索移動物體2 1 円争止物體3，並將移動式攝像機1〇的方向與倍率對準於言夕 50 本紙張尺度適用中國國豕標準（CNS)A4規格（210 X 297公g ---^----------裝------—訂—-------線 (請先閱讀背面之注意事項再填寫本頁) 540010 A7 B7 五、發明說明（4) 方向。且，位置/大小檢測機構17亦能自粗邊緣資訊112， 求出移動物體2及靜止物體3之領域數量的槪略大小。第 二，邊緣資訊形成機構15係使用邊緣資訊生成機構14所 生成之粗邊緣資訊圖像1Π，來生成形成邊緣資訊圖像115 。亦即，在邊緣資訊形成機構15生成形成邊緣資訊圖像 115之期間，移動物體2會產生移動。一般說來，由於移 動式攝像機10具有物理性機構，因此移動式攝像機10的 移動速度不會太快。因此，當該視覺裝置使移動式攝像機 10移動時，由於移動物體2的移動速度，而有來不及控制 移動式攝像機10的可能。由於上述原因，申請專利範圍第 4項之視覺裝置，未必需要移動物體2及靜止物體3之明 確且明瞭的形成邊緣資訊114。 其次，攝像機/環境座標轉換機構20及圖像/環境座標 轉換機構21，將用位置/大小檢測機構17所生成之重複資 訊圖像132所顯示之移動物體2及靜止物體3的領域位置 ，轉換爲環境座標系統的位置。此時，若框圖像1中移動 物體2及靜止物體3之領域位置的總數是2以上的話，環 境座標系統上亦存在兩個以上的位置。因此，爲了控制移 動式攝像機10之移動鏡頭、傾斜及變焦中任一項以使移動 式攝像機10的方向及倍率對準一個物體，必須從環境座標 系統上的位置中選擇一個。位置選擇機構22根據一定的判 斷基準，選擇環境座標系統上的一個位置。此處所使用之 判斷基準主要如下。第一，選擇環境座標系統上物體最近( 或最大的）者。此係因物體較遠(或較小時）時，邊緣資訊生 51 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐) --------·------^衣 (請先閱讀背面之注意事項再填寫本頁) 訂---------線· 經濟部智慧財產局員工消費合作社印制衣 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（h) 成機構14有可能生成雜訊，故盡量選擇環境座標系統上之 位置有物體之槪率較高者。第二，當環境座標系統上之一 定範圍內，複數之位置集中時，選擇其中之一。這有兩種 可能性。一是邊緣資訊生成機構14對一個物體生成分散的 粗邊緣資訊112的可能性，另一個是實際上存在複數個物 體的可能性。第三，環境座標系統上多數位置係大致相同 距離（或大致相同大小）時，選擇離原點最近的位置，亦即 ，選擇離移動式攝像機10方向最近的位置。視用途與狀況 適當組合上述事項，位置選擇機構22即能選擇環境座標系 統上的一個位置。 又，申請專利範圍第4項之視覺裝置，除了以位置選 擇機構22所選擇的位置外，亦能在下述位置求出搖動鏡頭 、傾斜及變焦移動式攝像機10。首先，該視覺裝置必須使 移動式攝像機10振動。因此，爲了使移動式攝像機10振 動，振動指令生成機構25，將移動式攝像機10移動之位 置指定爲環境座標系統上的位置。振動指令生成機構25所 指定之位置，係移動式攝像機10絕對不振動的範圍，而由 擬似隨機數等加以決定。再者，申請專利範圍第4項之視 覺裝置，係藉輸入來自外部之控制指令以求出對移動式攝 像機10進行搖動鏡頭、傾斜及變焦。至於一般的用途，由 於係對移動式攝像機10現在面向的方向與倍率，進行移動 鏡頭、傾斜及變焦，故使用控制指令輸入機構24暫時記憶 控制指令後，即算出將移動式攝像機10之現在位置作爲原 點之環境座標系統上的位置。又，藉改良控制指令輸入機 52 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） ---^------------------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（(r ) 構24，能容易地使移動式攝像機10移動至特定的位置。 因此，爲了控制移動式攝像機10的移動鏡頭、傾斜及 變焦，運動控制機構23自前述環境座標系統上的三個位置 中選擇一個位置。運動控制機構23自前述環境座標系統上 的三個位置中選擇一個位置時，以控制指令輸入機構24、 位置選擇機構22、振動指令生成機構25的順序，自有輸 入位置之機構加以選擇。所選擇的位置從環境座標系統轉 換爲攝像機座標系統。之後，利用攝像機指令生成機構26 ，轉換爲移動式攝像機10能辨識的指令，以傳送至移動式 攝像機10。據此，申請專利範圍第4項之視覺裝置，即能 控制移動式攝像機10的移動鏡頭、傾斜及變焦。 又，視移動式攝像機10之機種，若不傳送特定指令給 移動式攝像機10的話，就無法得到移動式攝像機1〇的狀 態’亦即，無法得到移動式攝像機1〇之方向及倍率、與移 動式攝像機10是在移動中還是停止中的資訊。因此，運動 控制機構23自三個位置中選擇一個並輸出至在攝像機指令 生成機構26後，指示攝像機指令生成機構26，以將詢問 移動式攝像機1〇是否移動中的資訊指令傳送給移動式攝像 機10 ’從移動式攝像機10透過攝像機/環境座標轉換機構 20 ’待機到接收該資訊爲止。若所接收的資訊係顯示在移 動中’即再一次指示攝像機指令生成機構26，以將詢問移 動式攝像機10是否移動中的資訊指令傳送給移動式攝像機 10右所接收的資日只係顯不停止中’即指不攝像機指令生 成機構26，以將詢問移動式攝像機10現在位置及倍率之 53 本紙張尺度適用中國國家標準（CNS)A4規格咖χ 297公爱） —1 —·--------------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（0) 指令傳送給移動式攝像機10。在此期間，運動控制機構23 不選擇三個位置。攝像機指令生成機構26根據來自運動控 制機構23的指令，將對應的指令傳送給移動式攝像機1〇 。攝像機/環境座標轉換機構20原封不動地將移動式攝像 機10是否移動中的資訊傳送給運動控制機構23，另外， 將移動式攝像機10的現在位置及倍率，自攝像機座標系統 轉換爲環境座標系統的位置。據此，申請專利範圍第4項 之視覺裝置，即能一面逐次調查移動式攝像機10的狀態， 一面控制移動式攝像機10的移動鏡頭、傾斜及變焦。 到現在爲止，已就具備有移動式攝像機10爲常時以適 當的大小拍攝框圖像1中之移動物體2及靜止物體3之基 本機構的申請專利範圍第4項之視覺裝置，進行了說明。 然而，並不表示移動物體2及靜止物體3是經常存在於移 動式攝像機10所拍攝的範圍中，且移動物體2不久後即會 自移動式攝像機10所拍攝的範圍移動到其他的位置。當然 ，若移動物體2移動，從移動式攝像機1〇所見之移動物體 2的形狀及顏色當然變化。因此，以下所說明之申請專利 範圍第5項之視覺裝置，可視爲在申請專利範圍第4項之 視覺裝置中，增加用來更正確地辨識框圖像1中之移動物 體2及靜止物體3的形狀及顏色的若干機構，與用來更正 確地計具移動式攝像機10在一'定時間內所拍攝之移動物體 2及靜止物體3之數量的若干機構，與更正確地使移動式 攝像機10移動，以朝向在移動式攝像機1〇能拍攝之範圍 內的移動物體2及靜止物體3之位置的機構。 54 本紙張尺度適用中國國家標準（CNS)A4規格C210 X 297公釐） --^ I L.---.------^^衣--------訂---------線 2清先閱讀背面之注音？事項再填寫本頁} 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（fj) 第7圖之視覺裝置，係在申請專利範圍第4項之視覺 裝置中追加邊緣資訊形成機構15、物體/背景分離機構16 、領域規格化機構27及規格化圖像保持機構28，以生成 規格化圖像145者。邊緣資訊形成機構15、物體/背景分離 機構16，與由移動物體部101及所有物體計數部1〇2所構 成之申請專利範圍第1項到第3項之視覺裝置的邊緣資訊 形成機構15、物體/背景分離機構16相同者。 當領域規格化機構27分別輸入來自物體/背景分離機 構16及圖像取得機構11之物體領域圖像142及數位圖像 111時，即從數位圖像111切出分離物體領域143，一面使 分離物體領域143變形，一面配合數位圖像111的圖像尺 寸，盡可能利用內插及放大生成規格化圖像145。由於領 域規格化機構27僅用各像素之附近處理，即能將分離物體 領域143進行規格化，因此適合並列化。若領域規格化機 構27能並列輸出規格化圖像145的全部像素，自領域規格 化機構27至規格化圖像保持機構28之通信，即能各像素 並列進行。 當規格化圖像保持機構28自領域規格化機構27輸入 規格化圖像145時，若規格化圖像145之輸出端要求適當 格式之規格化圖像145的話，即能將規格化圖像145轉換 爲規格化圖像145輸出端所要求的格式。之後，規格化圖 像保持機構28在將規格化圖像145確實地傳送至規格化圖 像145的輸出端前，記憶規格化圖像145 —定期間。規格 化圖像保持機構28若限定轉換格式的話，由於僅藉各像素 55 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） n —w n I 11«»11 n ϋ ϋ I * n ·ϋ n (請先閱讀背面之注意事項再填寫本頁) 訂---------線- 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（α) 附近處理即能轉換規格化圖像145，因此適合並列化。若 規格化圖像保持機構28能並列輸出規格化圖像145的全部 像素’自規格化圖像保持機構28至規格化圖像145之輸出 端的通信，集能各像素並列進行。 因此，藉使用邊緣資訊形成機構15 '物體/背景分離機 構16、領域規格化27及規格化圖像保持機構28，即使申 請專利範圍第4項之視覺裝置所拍攝的移動物體2及靜止 物體3，在框圖像1中的任意位置以任意的大小存在，第7 圖之視覺裝置亦能儘量生成類似之移動物體2及靜止物體 3的規格化圖像145。據此，當規格化圖像145之輸出端須 辨識移動物體2及靜止物體3時，規格化圖像145的輸出 端即不需考慮框圖像1中的移動物體2及靜止物體3的位 置及大小，而能使用適當之辨識方法。 又，第7圖之視覺裝置中，在邊緣資訊生成機構14、 邊緣資訊形成機構15、物體/背景分離機構16、領域規格 化27及規格化圖像保持機構28間，各機構所輸入之圖像 解析度或影像大小並不一定要一致。例如，第7圖之視覺 裝置，於邊緣資訊生成機構14中輸入降低數位圖像111之 解像度的低解像度數位圖像116，另一方面，於邊緣資訊 形成機構15中，輸入以適當之方法將邊緣資訊生成機構 14所生成的低解像度粗邊緣資訊圖像117之圖像尺寸放大 爲數位圖像111之圖像尺寸的粗邊緣資訊圖像113,於物 體/背景分離機構16及領域規格化機構27中輸入數位圖像 111，即能減低邊緣資訊生成機構14的負載。亦即，幾乎 56 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） ------1------裝--------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明（ 不需改變邊緣資訊形成機構15以後所生成之規格化圖像 145的品質，即能更高速地控制位置/大小檢測機構17以後 之移動式攝像機10的移動鏡頭、傾斜及變焦的控制。若進 一'步進行該方法，於邊緣資訊形成機構15中’輸入邊緣資 訊生成機構14所生成之粗邊緣資訊圖像117的圖像中切出 存在粗邊緣資訊112之領域的切出粗邊緣資訊圖像119, 於物體/背景分離機構16及領域規格化27中，輸入自數位 圖像111中切出與切出粗邊緣資訊圖像119相同位置領域 的數位圖像120，即能減低邊緣資訊形成機構15以後之規 格化圖像145所生成的負載。此處，若能利用申請專利範 圍第4項之視覺裝置，使移動式攝像機10在框圖像1之中 央以適當的大小拍攝移動物體2及靜止物體3的話，即能 預先決定對數位圖像111的切出粗邊緣資訊圖像119及切 出數位圖像120的切出領域。據此，第7圖之視覺裝置， 即能利用1台移動式攝像機10，達成與使用廣角攝像機與 高解像度攝像機的物體搜索裝置同樣的性能。 第8圖之視覺裝置，係在第7圖之視覺裝置中追加圖 像辨識機構29及辨識機構保持機構30，以生成辨識結果 者。 當圖像辨識機構29自領域規格化機構輸入規格化圖像 145時，使用適當之圖像辨識方法辨識規格化圖像145之 移動物體2及靜止物體3的規格領域144，並輸出辨識結 果。由於輸入圖像辨識機構29之規格化圖像145，係利用 領域規格化機構27使移動物體2及靜止物體3變形，因 57 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） — III — 1·1 I I I I ·1111111 « — — — — — — II Αν (請先閱讀背面之注意事項再填寫本頁) 540010 A7 B7______ 五、發明說明Lrt) (請先閱讀背面之注意事項再填寫本頁) 此圖像辨識機構29，與其使用行程抽出法、傅立葉轉換及 哈佛轉換等對位置偏離較佳之方法進行幾何解析，倒不如 進行輸入圖像與模板圖像之比較的圖像匹配方法較爲合適 。作爲圖像匹配方法，可使用以誤差反傳播法（back propagation)而能學習模板圖像之傳播等的中性網路。又， 當使用中性網路時，藉使用中性網路的加速器，也能進行 並列化及高速化。規格化圖像145的辨識結果’係自圖像 辨識機構29輸出到辨識結果保持機構30。 當辨識結果保持機構30自圖像辨識機構29輸入規格 化圖像145之辨識結果時，若辨識結果之輸出端要求適當 格式之信號的話，即將辨識結果轉換爲辨識結果輸出端所 要求的格式。之後，辨識結果保持機構30在能確實傳送辨 識結果至辨識結果輸出端前，記億辨識結果一定期間。 因此，藉使用圖像辨識機構29及辨識結9果保持機構 30，第8圖之視覺裝置能生成移動式攝像機1〇以適當之大 小拍攝之移動物體2及靜止物體3的辨識結果。據此，辨 識結果之輸出端，能利用第8圖之視覺裝置，作爲移動式 攝像機10所拍攝之移動物體2及靜止物體3的辨識裝置。 經濟部智慧財產局員工消費合作社印制衣 第9圖之視覺裝置，係在第8圖之視覺裝置增加環境 理解機構31、計時機構32及環境地圖機構保持33，以生 成環境地圖。 計時機構32係利用定時器電路以1釐米(mm)秒單位輸 出現在時刻。現在時刻不斷地從計時機構32輸出至環境理 解機構31。 58 ‘本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公爱) — " " 540010 A7 B7 五、發明說明（q) 當環境理解機構31自圖像辨識機構29輸入移動物體 2及靜止物體3的辨識結果時，即作成由辨識結果、移動 式攝像機10在環境座標系統上的位置及現在時刻所構成的 環境資料。又，當環境理解機構31自圖像/環境座標轉換 機構21，輸入框圖像1中全部之移動物體2及靜止物體3 之領域在環境座標系統上之位置時，即僅作成移動物體2 及靜止物體3之領域數量的環境資料，該環境資料係由無 效資料所構成的辨識結果、移動式攝像機1〇在環境座標系 統上的位置加上框圖像1中之一個移動物體2及靜止物體 3之領域在環境座標系統上之位置的位置，以及現在時刻 所構成。此時，環境地圖係較現在時刻一定時間前所作成 之環境資料的集合，環境資料中的位置係以將移動式攝像 機10之原點作爲原點的環境座標系統來表示。環境理解機 構31 ’係隨時間之經過於對環境地圖追加及削除環境資料 。此外，環境理解機構31，在削除重複之內容的環境資料 的同時，當辨識結果是無效資料的環境資料中的位置在辨 識結果不是無效資料的其他環境資料中的位置附近時，削 除辨識結果爲無效資料的環境資料。再者，當辨識結果不 是無效資料之環境資料中的位置，係在辨識結果不是無效 資料之其他環境資料中的位置附近時，若該等環境資料中 的辨識結果一致的話，即削除前者的環境資料，根據環境 資料之記錄時間與環境資料中位置附近的範圍，決定環境 地圖的精度。環境地圖係從環境理解機構31輸出到環境地 圖保持機構33。 59 本紙張尺度適用中國國家標準（CNS)A4規格（21〇 x 297公釐） --------------，罐衣 (請先閱讀背面之注意事項再填寫本頁) ----訂---------線· 經濟部智慧財產局員工消費合作社印製 經濟部智慧財產局員工消費合作杜印制衣 540010 A7 B7 五、發明說明（㈡?） 當環境地圖保持機構33從環境理解機構31輸入環境 地圖時，若環境地圖之輸出端要求適當格式之信號的話， 即將環境地圖轉換爲環境地圖之輸出端所要求的格式。然 後，環境地圖保持機構33，在將環境地圖確實地傳送至環 境地圖之輸出端前，記憶環境地圖一定時間。 因此，藉使用環境理解機構31、計時機構32及環境 地圖保持機構33，第9圖之視覺裝置，能特定移動式攝像 機10所拍攝之移動物體2及靜止物體3的位置。據此，環 境地圖之輸出端，能利用第9圖之視覺裝置，作爲移動式 攝像機10所能拍攝之移動物體2及靜止物體3之位置的特 定裝置。 第10圖之視覺裝置，係在第9圖之視覺裝置中增加物 體位置推算機構，以將移動式攝像機10控制於物體之推算 位置。 當物體位置推算機構34從環境理解機構31輸入環境 地圖時，即選擇一個辨識結果爲無效資料的環境資料，抽 出該環境資料中的位置。從該位置減去利用攝像機/環境座 標轉換機構20所計算之移動式攝像機1〇的環境座標系統 上的現在位置’物體位置推算機構34，即能求出將移動式 攝像機10之環境座標系統上的現在位置作爲原點，即使過 去已生成粗邊緣資訊112,但辨識結果尙未出現之移動物 體2及靜止物體3在環境座標系統上的相對位置。但是， 於移動物體2之情形中，於現在該位置上不一定存在移動 物體2。因此，該位置係有可能存在移動物體2及靜止物 60 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） ---Γ---*------^^裝--------訂---I-----線 (請先閱讀背面之注意事項再填寫本頁) 540010 A7 B7 五、發明說明（4) 體3之移動物體2及靜止物體3的推算位置。另一方面， 若環境地圖中沒有辨識結果爲無效資料之環境資料的話， 即在移動式攝像機10能移動之範圍內生成適當之環境座標 系統上的位置。此時，作爲生成適當位置的基準，考率如 下項目。第一，利用擬似隨機數，生成任意的位置。第二 ，計算環境地圖中之環境資料中的位置密度低的位置。第 三，以適當之順序依序生成移動式攝像機10能移動之範圍 內的位置。例如，於移動式攝像機10可移動之範圍內，從 左上端的位置向右依序生成位置，到達右端後向下降一段 ，之後，向左依序生成位置，到達左端後向下降一段，之 後，向右依序生成位置，反覆此種動作。視用途與狀況組 合該等基準，即能有效地推算移動式攝像機10未拍攝之移 動物體2及靜止物體3的推算位置。物體推算位置係從物 體位置推算機構34，輸出到運動控制機構23。 此處，運動控制機構23是以申請專利範圍第4項之視 覺裝置的運動控制機構23爲基準，作如下之變更。亦即， 爲了控制移動式攝像機10的移動鏡頭、傾斜及變焦，運動 控制機構23是用控制指令輸入機構24、物體位置推算機 構34、位置選擇機構22及振動指令生成機構25的順序， 從有輸入位置之機構加以選擇。但是，運動控制機構23必 須不連續選擇來自物體位置推算機構34的輸入位置。 因此，藉使用物體位置推算機構34，第10圖之視覺 裝置，即能搜索存在於移動式攝像機10能拍攝範圍的移動 物體2及靜止物體3。據此，環境地圖之輸出端’能利用 61 本紙張尺度適用中國國家標準（CNS)A4規格（210 x 297公釐） -------J------ (請先閱讀背面之注意事項再填寫本頁) 訂------ 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 經濟部智慧財產局員工消費合作社印製 五、發明說明（I6) 第10圖之視覺裝置.，作爲移動式攝像機10所能拍攝之移 動物體2及靜止物體3之位置的特定裝置。 第11圖之視覺裝置，係在第10圖之視覺裝置中增加 物體計數機構35及物體數保持機構36，以生成任意物體 數。 當物體§十數機構35從環境理解機構31輸入環境地圖 時’即能在環境地圖中計算具有特定物體意義之辨識結果 的環境資料數，以生成任意物體數。環境資料中之辨識結 果，可從圖像辨識機構29所辨識的物體種類中選擇任意數 量。任意物體數是從物體計數機構35輸出到物體數保持機 構36。又’若視用途而有需要時，藉變更物體計數機構35 ，就能容易指定應從外部計數的物體種類。 當物體數保持機構36從物體數機構35輸入任意物體 數時’在將任意物體數確實地傳送至任意物體數之輸出端 前，記億任意物體數一定時間。 因此，藉使用物體計數機構35及物體數保持機構36 ’第Π圖之視覺裝置，能計算存在於移動式攝像機10能 拍攝範圍內之特定移動物體2及靜止物體3的數量。據此 ’任意物體數輸出端，能利用第11圖之視覺裝置，作爲移 動式攝像機10所能拍攝之移動物體2及靜止物體3之位置 的特定裝置。 如第12圖所示，申請專利範圍第5項之視覺裝置，係 在第11圖之視覺裝置中增加幾何解析機構37，以正確且 局速作成環境地圖。 62 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） (請先閱讀背面之注意事項再填寫本頁) -丨線- 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（以） 當幾何解析機構37從邊緣資訊形成機構15輸入形成 邊緣資訊圖像115時，即藉進行行程抽出法、傅立葉轉換 及哈佛轉換等幾何解析，推算框圖像1中的移動物體2及 靜止物體3的形狀，以生成幾何解析結果。而以往的圖像 辨識，係從圖像本身直接抽出線部分，或使用不影響位置 與大小等的各種轉換，進行辨識圖像中的物體。因此’雖 然處理非常單純，但組合的數量龐大，其結果需要龐大的 計算量，在得到所希望的結果前花費長的時間，或只能得 到低品質的辨識結果。然而，申請專利範圍第5項之視覺 裝置，是對框圖像1中之移動物體2及靜止物體3的圖形 辨識，由物體/背景分離機構16、領域規格化機構27及圖 像辨識機構29加以進行。因此，幾何解析機構37使用形 成邊緣資訊圖像115，近對物體/背景分離機構16、領域規 格化機構27及圖像認識機構29所難以應付的移動物體2 及靜止物體3的輪廓等圖形，以用簡單的方法大致進行解 析，即能以圖像辨識機構29省略不必要的圖形匹配，以環 境理解機構31正確且高速作成環境地圖，同時也能減低幾 何解析機構37本身的負載。幾何解析結果是從幾何解析機 構37，輸出到圖像辨識機構29及環境理解機構31。因此 ，圖像辨識機構29及環境理解機構31，分別進行如下的 動作。 當圖像辨識機構29從幾何解析機構37輸入幾何解析 結果時’首先，判斷是否爲圖像辨識機構29所辨識之對象 的圖形。若幾何解析結果是對象外圖形的話，圖像辨識機 63 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） ------------裝--------訂--------線 (請先閱讀背面之注意事項再填寫本頁) 540010 A7 B7 五、發明說明（tl) 構29即不動作。若幾何解析結果是對象圖形的話，圖像辨 識機構29即使用對象圖形用的模板圖像，進行圖形匹配。 例如’爲以誤差逆傳播法進行學習之視感控制器 (perxeptron)時，對每一對象圖形進行視感控制器之學習， 以作成學習資料，然後利用幾何解析結果，選擇學習資料 ’針對對象圖形有效地進行圖形匹配。亦即，若幾何解析 結果是圓形的話，即僅利用人臉與球等圓形物體之學習資 料’若幾何解析結果是三角形的話，即僅利用鐵路標誌與 類似山等三角形物體之學習資料。如此，不僅能利用小型 的視感控制器辨識多數物體之圖形，視感控制器本身也能 有效地學習。 當環境理解機構31從幾何解析機構37輸入幾何解析 結果時，首先，判斷是否爲圖像辨識機構29所辨識之對象 的圖形。若幾何解析結果是對象外圖形的話，環境理解機 構31即立刻從環境地圖中削除環境資料(係位於移動式攝 像機10之環境座標系統上的位置附近，辨識結果爲無效資 料的環境資金）。據此，能從環境地圖中削除不必要的環境 資料，即使不輸出物體位置推算機構34等不必要的物體推 算位置亦可。當幾何解析結果是對象圖形時，即待機到從 圖像辨識機構29輸入辨識結果。 因此，藉使用幾何解析機構37，申請專利範圍第5項 之視覺裝置，能正確且高速計算存在於移動式攝像機1〇能 拍攝範圍內之特定移動物體2及靜止物體3的數量。據此 ，任意物體數的輸出端，即能利用申請專利範圍第5項之 64 -------------- (請先閱讀背面之注意事項再填寫本頁) 訂---------線 經濟部智慧財產局員工消費合作社印製 本紙張尺度適用中國國家標準（CNS)A4規格(210 x 297公釐） 540010 A7 B7 五、發明說明（Μ) 視覺裝置，作爲高速計算移動式攝像機10能拍攝之特定移 動物體2及靜止物體3之數量的裝置。 (請先閱讀背面之注意事項再填寫本頁) 另外，從申請專利範圍第1項到第5項之視覺裝置， 所使用的圖像記憶機構12、圖像振動機構13、邊緣資訊生 成機構14、邊緣資訊形成機構15、物體/背景分離機構16 、位置/大小檢測機構Π、領域規格化機構27、規格化圖 像保持機構28及圖像辨識機構29能使用由陣列運算單元 40(Array Operation Unit)所構成的資料處理裝置no，進行 組裝。因此，以下以利用陣列運算單元40之資料處理裝置 110的實施形態爲例，參照圖面說明從申請專利範圍第6 項到第12項之視覺裝置。 經濟部智慧財產局員工消費合作社印製 首先，陣列運算單元40，使用輸入圖像的一個像素與 其附近像素，生成輸出圖像的一個像素。因此，如第13圖 所示，使陣列運算單元40配合輸入圖像的大小，使用網格 狀排列的資料處理裝置110，資料處理裝置110即能從輸 入圖像生成輸出圖像。又，第13圖中，將陣列運算單元 40簡寫爲A0U。其次，陣列運算單元40可利用專用硬體 來進行組裝，亦可在一般電腦上利用軟體來進行組裝。亦 即，若能從輸入圖像生成輸出圖像的話，組裝機構不受限 制。因此，藉顯示陣列運算單元40的算法，能顯示資料處 理裝置110的圖像處理。因此，爲了顯示陣列運算單元 的算法，就圖像記憶機構12(參照第1圖及第6圖）、圖像 振動機構13(參照第3圖、第4圖及第5圖）、邊緣資訊生 成機構14(參照第1圖及第6圖）、邊緣資訊形成機構ι5(參 65 適 度 尺 張 紙 本 格 I規 4 )A S) N (C 準 標 540010 Α7 Β7 五、發明說明) (請先閱讀背面之注意事項再填寫本頁) 照第1〜5圖、第7〜12圖）、物體/背景分離機構16(參照第 2、4、5、7〜12圖）、位置/大小檢測機構17(參照第1圖及 第6圖）、領域規格化機構27(參照第7圖）、規格化圖像保 持機構28(參照第7圖)及圖像辨識機構29(參照第8圖)所 -使用的式，加以說明。 設寬度w、高度h、頻帶數b的任意2n色調爲X、y、 w，X、y、W是使用各個位置p(i、j、k)的頻帶像素値Xuk、 yuk、wuk，如式1、式2、式3所顯示。還有，粗體字是顯 示向量。另外，11爲非負的整數，〜、}1、13、卜」、]^爲自 然數。 X = is value at k),l<i<w,l<j<h,l<k<b] (1) y = {Vijk\Vijk is value at k),l < i < w,l < j < h,l < k < b} (2) w = {y^ijk\^ijk is value at p(z, <i <w,l < j <h,l <k <b} (3) 首先，對前述圖像的各頻帶像素値，就有關點處理的 函數，說明如下。 當圖像X轉換爲二値圖像時，根據式4，將頻帶像素 値加以二値化。 經濟部智慧財產局員工消費合作社印制衣 ΦύΑ:(χ) ⑷ 1 if xijk > Ο, L 0 otherwise. 當將圖像x轉換爲頻帶最大値圖像時，根據式5，從 行j列像素的各頻帶値中，選擇最大値。又，由於前述頻 帶最大値圖像爲單頻帶圖像，故爲了方便，將之作爲頻帯 數1之前述圖像加以處理。因此，函數BuKx)的第3附加 66 本紙張尺度適用中國國家標準（CNS)A4規格松10 X 297公釐） 540010 經濟部智慧財產局員工消費合作社印製 A7 五、發明說明（β) 字爲1。 ^71 (x) = max{xijk} ⑸ k 圖像x舄二値圖像，當反轉圖像χ時，根據式6加以 計算。 ^ijk{x) = 1 — Xijk (6) 圖像χ的位置P(1、j、k)中的對數轉換是根據式7進行 。在此式中，e爲偏置，係爲了使自然對數函數的輸出値 在有效範圍而加以使用，一般說來，e==1即可。利用此對 數化即能將頻帶像素彼此相除變爲相減。另外，設圖像X 爲2n色調的數位圖像時，與頻帶數無關的，在記憶體上若 具有包含2n個要素的一覽表的話，即不需每次計算自然對 數函數’亦不需具有標準的對數表。Order --------- Line I Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 540010 Α7 Β7 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 5. Description of the invention (3ί) The aforementioned parameters are easily revised. In addition, the array operation unit does not necessarily need to wait closely for the reception of the nearby images of the various images transmitted from the nearby arrangement device. This is because when the aforementioned nearby pixels of various aforementioned images cannot be received from the aforementioned array operation unit located nearby, the aforementioned array operation unit waiting to be received can be replaced by its own parameter 値 or pixel 取代. At this time, although there may be some noise in the pixels of the various images generated by the array operation unit, in the present invention, almost all noises are absorbed by the various mechanisms described above. By using the aforementioned parameter 値 or the aforementioned band pixel 値 to replace it, the edge processing and timeout processing can be realized simultaneously and simply. The invention in the thirteenth aspect of the patent application is a visual device, which is characterized by having a data input mechanism, a mechanism for sequentially storing the aforementioned data, a mechanism for transmitting the aforementioned data between array computing units, and a calculation using the aforementioned data. The array computing unit of the mechanism and the mechanism that outputs the aforementioned data includes: a mechanism that arranges the array computing unit in a grid shape; a mechanism that combines nearby array computing units with each other according to the respective positional relationship of the array computing unit; A mechanism for data communication between the adjacent array operation units; and a mechanism for independently operating each of the array operation units. Regarding the aforementioned institutions that input the aforementioned data, the aforementioned institutions that sequentially memorize the aforementioned data, the aforementioned institutions that transmit the aforementioned data between the aforementioned array computing units, the aforementioned institutions that use the aforementioned data to perform calculations, and the aforementioned assets are output. Standard (CNS) A4 size 4210 X 297 mm) —. —, -------------- Order --------- line (Please read the notes on the back before filling this page) 540010 A7 _____B7 _ V. Description of the invention (P ) (Please read the precautions on the back before filling out this page.) Regarding the aforementioned mechanism of the material, each of the aforementioned array operation units has the same operation steps regardless of the position of its grid arrangement. Therefore, 'when hardware is used to assemble the aforementioned array computing unit', the same circuits that implement the aforementioned array computing unit can be arranged on the plane with correct rules' Since these circuits only need to be connected to neighbors, wiring The amount is also small, as long as the size of the processed image is increased or decreased, and the circuits can be operated in parallel. When assembling the array operation unit using software, a visual device that grids the array operation unit in parallel can be executed using a program having a high degree of parallelism. The invention of claim 14 in the scope of patent application is a visual device, which is characterized in that: the array operation unit is provided with a processor having an input data processing mechanism, a memory for storing the processing programs and variables of the foregoing data, and a memory for A controller for communicating with an adjacent array operation unit; the controller is provided with a mechanism for storing the inputted data in the memory, and a mechanism for transmitting the variables in the memory to the array operation unit, And the machine that memorizes the variables transmitted by the adjacent array operation units in the aforementioned memory is printed by the consumer cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs, that is, the circuits of the array operation units are assembled by hardware. The array operation unit may use a general processor for processing the aforementioned data input, and a general memory for storing programs and variables for processing the aforementioned data. The controller, when the array operation unit is combined with a maximum of around 4, can only transmit variables to the adjacent array operation unit, but when the array operation unit and each of the vicinity of 8 are above 34 paper sizes Applicable to China National Standard (CNS) A4 specification (, 210 X 297 mm) 540010 Printed by A7 B7, Consumer Cooperatives, Intellectual Property Bureau of the Ministry of Economic Affairs 5. Description of the invention The aforementioned variables of are temporarily transmitted to the adjacent array operation unit, so the aforementioned variables are transmitted to themselves, and then they can be received. In addition, by using the aforementioned mechanism that transfers each of the variables in the memory to the adjacent array operation unit, it is also possible to transfer its own variable to the array operation unit that is not included in the vicinity of 4. According to this, as far as the hardware device is concerned, although the array operation unit is only connected to the adjacent array operation unit, it can also properly communicate the aforementioned data with the array operation unit near 8 or more. In addition, it is possible to appropriately solve various problems related to the hardware assembly and the actual processing among the problems described above. [Brief description of the drawings] Fig. 1 is a block diagram of a visual device for calculating the number of moving objects from a frame image. Figure 2 is a block diagram of a visual device that uses an object / background separation mechanism to calculate the number of moving objects from a frame image. Figure 3 is a block diagram of a visual device that calculates the number of all objects from a frame image. Figure 4 is a block diagram of a visual device that uses the object / background separation mechanism to calculate the total number of objects from the frame image. Figure 5 is a block diagram of a visual device used to determine the ratio of moving objects to stationary objects. Figure 6 is a block diagram of a vision device used to control a mobile camera. Figure 7 is a block diagram of a vision device that generates a normalized image of an object. 35 This paper size is applicable to China National Standard (CNS) A4 (210 X 297 mm) ---------------------- Order ------- -Line (Please read the precautions on the back before filling this page) 540010 A7 B7 V. Description of the Invention Figure 8 is a block diagram of a visual device that outputs the result of object recognition. Brother 9 is a block diagram of a visual device that generates a map of the environment. Fig. 10 is a block diagram of a vision device for controlling a mobile camera using an environment map. Figure 11 is a block diagram of a visual device that counts any number of objects. Fig. 12 is a block diagram of a high-speed vision device using geometric analysis. Fig. 13 is a block diagram of an array operation unit in a grid configuration. Fig. 14 is a flowchart showing the algorithm of the image memory mechanism of this embodiment. Fig. 15 is a flowchart showing the algorithm of the image vibration mechanism of this embodiment. Fig. 16 is a flowchart showing the algorithm of the edge information generating mechanism in this embodiment. Fig. 17 is an explanatory diagram when forming rough edge information using digital images to form edge information. Fig. 18 is a flowchart showing the algorithm of the edge information forming mechanism in this embodiment. FIG. 19 is an explanatory diagram when forming low-resolution coarse edge information generated from a low-resolution digital image into edge information. FIG. 20 is an explanatory diagram when a low-resolution coarse edge information area generated from a low-resolution digital image is cut out and formed into edge information. Fig. 21 is an explanatory diagram when detecting the position and size of an object in an edge information image. 36 (Please read the precautions on the back before filling out this page) | Install -------- Order --------- Printed by the Consumers' Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs This paper is suitable for China National Standard (CNS) A4 (210 X 297 mm) Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economy 540010 A7 B7 V. Description of the Invention Figure 22 is a flowchart showing the algorithm of the position / size detection mechanism of this embodiment. . Fig. 23 is an explanatory diagram when detecting the position and size of an object in an image of an object area. Fig. 24 is an explanatory diagram when a separated object area of a digital image is normalized. Fig. 25 is a flowchart showing the algorithm of the domain normalization mechanism in this embodiment. Fig. 26 is a flowchart showing the algorithm of the normalized image holding mechanism in this embodiment. Fig. 27 is an explanatory diagram when matching graphics from a template image with respect to a normalized image. Fig. 28 is a flowchart showing an algorithm of pattern matching in the image recognition mechanism of this embodiment. Fig. 29 is an explanatory diagram showing a state in which the edge information of the triangle is separated into the inside area and the outside area of the triangle. Fig. 30 is a flowchart showing the algorithm of the object / background separation mechanism in this embodiment. Fig. 31 is an explanatory diagram showing the state of the edge information of the triangle in the broken line state being separated into the inner area and the outer test area of the broken line triangle. Fig. 32 is an explanatory diagram showing a state in which edge information of two triangles overlapping is separated into two triangle areas and a background area. Fig. 33 is an explanatory diagram showing the state where the edge information of two circular dotted lines is separated into two circular areas and a background area. 37 Jade paper scale is applicable to Chinese National Standard (CNS) A4 specification 4210 X 297 mm 1 1 ------------- ΦΜί ----- 丨 Order --------- (Please read the precautions on the back before filling this page) 540010 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7 V. Description of Invention (#) Figure 34 is a block diagram of the internal structure of the array computing unit. Figure 35 is a block diagram of the controller. Fig. 36 is an explanatory diagram showing input and output signals of a mark decoder. Fig. 37 is an explanatory diagram showing input and output signals of a mark encoder. Fig. 38 is a flowchart showing an algorithm of a processor transmitting data to an adjacent array operation unit through a controller. Fig. 39 is a flowchart showing an algorithm for the controller to receive data from an adjacent array operation unit. Figure 40 is a flowchart showing the algorithm of the processor receiving data from the input register. [Best Embodiment of the Invention] Hereinafter, embodiments of the twelve visual devices shown in Figs. 1 to 12 will be described. These embodiments use an image acquisition mechanism 11 (refer to FIG. 1), an image memory mechanism 12 (refer to FIG. 1), an image vibration mechanism 13 (refer to FIG. 3), and an edge information generation mechanism 14 (refer to FIG. 1 and 3), edge information forming mechanism 15 (refer to FIG. 1), object / background separation mechanism 16 (refer to FIG. 2), position / size detection mechanism 17 (refer to FIGS. 1 and 2), The pixel counting mechanism 18 (see FIG. 1), the pixel holding mechanism 9 (see FIGS. 1 and 5), and the like. The image acquisition mechanism 11 is to receive a camera signal (frame image 1) and convert it into a digital image 111 of an appropriate format and size; the image memory mechanism 12 is to memorize the digital image 丨 11 for a certain period of time; The image vibration mechanism 13 uses a digital circuit to generate a digital image in vibration; the edge information generation mechanism 14 generates coarse edge information 112 of the moving object 2 or the stationary object 3 based on the two digital images in; the edge information is formed 38 ---------------------- Order --------- line (Please read the precautions on the back before filling this page) This paper size Applicable to China National Standard (CNS) A4 specification 4210 x 297 mm) ——Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 540010 Α7 ----- Β7 V. Description of the invention (q) Institution 15 is the rough edge information U2 It is formed into more specific formation edge information 114; the object / background separation mechanism 16 separates the area distinguished by the formation edge information U4; the position / size detection mechanism 17 detects each area distinguished or separated by the formation edge information 114 Position and size; pixel counting mechanism 18 Position, calculates the number of regions; holding means 19 is a pixel area ratio of the number of output or the number of regions. As shown in FIG. 1, the moving object counting unit 101 inputs the framed image 1 of the motion captured by the camera into the image acquisition mechanism 11, and forms it according to the image memory mechanism 12, the edge information generation mechanism 14, and the edge information. The mechanism 15, the position / size detection mechanism 17, the pixel counting mechanism 18, and the pixel holding mechanism 19 sequentially perform input and output, and output the number of the moving objects 2 reflected on the frame image 1. The frame image 1 may be a still image continuously captured by a digital camera. When the image acquisition mechanism 11 inputs a frame image 1 of a moving image from a camera, if the motion image is an analog signal, the frame image 1 is converted into a digital signal by A / D conversion using a general portrait frame acquisition board. To form a digital image 111. In addition, if the voltage of the CCD imaging element can be directly input, it can also be converted into a digital signal of an appropriate number of bits by A / D conversion. When a moving image is a digital signal, if it is compressed, it is expanded, and if it is not compressed, it is input as it is. With this method, since an arbitrary frame image 1 in a moving image can be cut out, the frame image 1 is cut out to form a digital image 111. Since the converted digital image 111 will have an arbitrary image size according to the appropriate format, the image acquisition mechanism cannot use pixel units to convert the image data into a format that can be referred to, and count 39 papers for moving objects. Standards are applicable to China National Standard (CNS) A4 specifications (210 X 297 mm) --1! ----------------- Order --------- line ( Please read the notes on the back before filling this page) 540010 A7 B7 Printed by the Consumers' Cooperative of Intellectual Property Bureau of the Ministry of Economic Affairs 5. Description of the Invention (4) Section 101 cuts out the required image size and outputs it as a digital image 111 . If the image acquisition unit 11 can output all the pixels of the digital image 111 in parallel, communication from the image acquisition unit 11 to the image memory unit 12 can be performed in parallel for each pixel. When the image memory mechanism 12 inputs the digital image 111 from the image acquisition mechanism 11, the digital image 111 is memorized for a certain period of time in accordance with the time-resolving ability of the moving object counting section 101 or the computing ability of each mechanism. That is, since the image storage mechanism 12 does not change the stored image even if the digital image 111 is input during this certain period of time, subsequent mechanisms can input the same digital image 111 at different points in time. Furthermore, the image memory mechanism 12 does not perform image processing on the digital image 111, so all pixels of the digital image 111 are memorized while maintaining the phase relationship of the second element. If the image memory mechanism 12 can output all the pixels of the digital image 111 in parallel, the communication from the image memory mechanism 12 to the edge information generating mechanism 14 can be performed in parallel for each pixel. When the edge information generating mechanism 14 inputs the digital image 111 from the image memory 12, it compares the previously input digital image 111 to generate the coarse edge information image 113 of the moving body 2. Since the edge information generating mechanism 14 can generate the coarse edge information image 113 only when processing is performed in the vicinity of each pixel, it is suitable for parallelization. If the edge information generating mechanism 14 can output all the pixels of the thick edge information image 113 in parallel, the communication from the edge information generating mechanism 14 to the edge information forming mechanism 15 can be performed in parallel with each pixel. When the edge information forming mechanism 15 inputs the coarse edge information image 113 from the edge information generating mechanism 14, it refers to the number 40 --- T stored in the image memory mechanism 12. (Please read the precautions on the back before filling out this page.) Γ This paper size is applicable to China National Standard (CNS) A4 (210 X 297 mm) 540010 A7 B7 V. Description of the invention (4) (Please read the back Note that you need to fill in this page again.) The bit image 111 generates the edge information image 115 of the moving object 2 which is more clear and clear than the digital image 113. Since the edge information formation mechanism 15 can process and generate the edge information image 15 ′ only in the vicinity of each pixel, it is suitable for parallelization. If the edge information forming mechanism 15 can output all the pixels forming the edge information image 115 in parallel, the communication from the edge information forming mechanism 15 to the position / size detection mechanism 17 can be performed in parallel for each pixel. When the position / size detection mechanism Π is input from the edge information forming mechanism 15 to form the edge information image 115, the position and size of the area of the moving object 2 indicated by the edge information 114 are detected. Since the position / size detection mechanism 17 processes only the vicinity of each pixel, that is, it can generate a repeated information image 132 showing the detection result of the position and size of the moving object 2 in the area, so it is suitable for parallelization. If the position / size detection mechanism Π can output all pixels of the repeated information image 132 in parallel, communication from the position / size detection mechanism Π to the pixel counting mechanism 18 can perform each pixel in parallel. Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. When the pixel counting mechanism 18 inputs the repeated information image 132 from the position / size detection mechanism 17, it can calculate the number of pixels in the area showing the moving object 2 of an appropriate size. Rather than using parallel processing, this processing is more convenient than sequential processing. Therefore, each pixel of the repeated information image 132 is stored in the linear memory. The number of pixels showing the position of the area of the moving object 2 is output from the pixel counting mechanism 18 to the pixel number holding mechanism 19. When the pixel number holding mechanism 19 inputs the number of pixels from the pixel counting mechanism 18, it converts the number of pixels to two's complement representation or floating decimal point representation, and outputs the necessary period. That is, 41 paper standards are applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) 540010 Printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7___ 5. Description of the invention (蛉.) Can output the number of moving objects. The aforementioned mechanisms can be implemented in parallel if they can be assembled with independent hardware, so that the moving object counting unit 101 can be instantaneous. Therefore, it is suitable for counting the number of fast-moving objects or processing a large number of moving images in a short time. In addition, if several organizations can use software to assemble on an I processor, the calculation speed is slower, but it can be manufactured cheaply. Therefore, it is very suitable for an application where the count result takes several minutes or more. When the position / size detection mechanism 17 uses the edge information formation image 115 generated by the edge information formation mechanism 15 to detect the position and size of the area of the moving object 2, the area of the moving object 2 may be displayed due to the density of the moving object 2. The number of pixels at the position is different from the number of moving objects 2. The main reason is that the formation of the edge information 114 does not necessarily mean that the edges of the moving object 2 are correctly grasped, and that the position / size detection mechanism 17 cannot discriminate the shape of the moving object 2 from the formation of the edge information 114. Therefore, if the density of the moving object 2 is high, the edge information 114 generated by different moving objects 2 will be combined, and sometimes it will be mixed with non-existing object edge information. In order to solve this problem, although the shape of the moving object 2 can be discriminated based on the formation of the edge information 114, the conventional geometric analysis method requires a large area for processing, so the amount of calculation is increased. If you want to improve the accuracy of the discrimination result, the calculation time That is, it becomes longer with an exponential function. Therefore, using the nearby processing, an object / background separation mechanism 16 (see Fig. 2) can be used as a mechanism to solve this problem. When the object / background separation mechanism 16 enters 42 from the edge information forming mechanism 15, the paper size is applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) —1 — * ------------ --Order --------- line (please read the notes on the back before filling this page) Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 540010 A7 ------- B7 V. Description of Invention (Ί ^ Ι) When the edge information image 115 is formed, the pixels included in the object area 141 and the pixels included in the background can be separated into different groups, and the separation results are sequentially output in group units. When the object areas 141 can be clearly distinguished by the edge information 114 even if they are adjacent, the object / background separation mechanism 16 'can separate these object areas into different groups. Therefore, the number of groups also reached three or more. Since the object / background separation mechanism 16 can separate the object area 141 and the background in each pixel only by the nearby processing, it is suitable for parallelization. If the object / background separation mechanism 16 can output all pixels of the object area image 142 in parallel, communication from the object / background separation mechanism 16 to the position / size detection mechanism 17 can be performed in parallel. Therefore, as shown in FIG. 2, by using the object / background separation mechanism 16, the edge information 114 can be supplemented by only the nearby processing while separating the area of the moving object 2 from other background areas. At this time, if the area of the moving object 2 is similar to a circle, in other words, if the area of the moving object 2 has less unevenness, the position / size detection mechanism 17 can grasp a certain size area to a specific position. Therefore, if it is known in advance that the moving object 2 is similar to a circle, the pixel counting mechanism 18 can use the object / background separation mechanism 16 to calculate the number of pixels that display the position of the moving object 2 in a specific size. In other words, the moving object counting section 101 can output a more accurate number of moving objects. Hereinabove, it has been described how the moving object counting section 101 operates when the number of moving objects 2 in the two or more frame images 1 of the moving image is counted. Next, the total number of moving objects 2 and stationary objects 3 in a frame image 1 of a dynamic image will be explained, that is, the feelings when calculating the number of all objects. 43 This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm) --------, -------------- Order --------- line (Please read the notes on the back before filling in this Page) 540010 A7 ------- B7 V. Description of Invention (Private) Form. / First, the basic operation is the same as that in the case where the number of moving objects 2 is calculated using the frame image 1 of the aforementioned moving image. That is, it is sufficient to replace the stationary object 3 in the frame image 1 with the moving object 2 in the frame image 1 by an appropriate method. In this way, since the rough edge information 112 of the stationary object 3 is generated, the moving object counting section 101 can also calculate the number of stationary objects 3. The simplest method of treating the stationary object 3 as the moving object 2 is to use a shaker or the like to finely oscillate a video camera (digital camera) or the stationary object 3 itself. However, since this method requires a physical mechanism, the moving body counting unit 101 becomes expensive or complicated. On the other hand, as a mechanism that does not use a physical mechanism, there is an image vibration mechanism 13 (see FIG. 3). When the image vibration mechanism 13 inputs a digital image 111 from the image memory mechanism 12, the stationary object 3 is at The digital image 111 vibrates up, down, left, and right in a range of about 3 pixels in an image unit or in a unit of a pixel. If the image vibration mechanism 13 can output all the pixels of the digital image 111 in parallel, the communication from the image vibration mechanism 13 to the edge information generation mechanism 14 can be performed in parallel with each pixel. Therefore, as shown in FIG. 3, the image vibrating mechanism 13 vibrates up, down, left, and right in the range of the three pixels before and after the stationary object 3 in the digital image 111 generated by the image obtaining mechanism 11. Accordingly, since the edge information generating mechanism 14 can treat the stationary object 3 as the moving object 2 to generate the coarse edge information 112 of the stationary object 3, all the object counting units 102 can calculate the moving object 2 and the stationary object 3 The total number, that is, can calculate all 44 paper sizes applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) —. — · ------ ΛΜ, (Please read the notes on the back before filling this page) Order --------- line «Printed by the Intellectual Property Bureau of the Ministry of Economic Affairs Consumer Cooperatives 540010 Intellectual Property of the Ministry of Economic Affairs A7 printed by the Bureau's Consumer Cooperatives V. Invention Description (Yu)) Number of objects. Furthermore, regardless of the moving object 2 and the stationary object 3, the object / background separation mechanism 16 separates the object area 141 and the background area using only the edge information U4. Therefore, if it can be known in advance that the moving object 2 and the stationary object 3 are similar to a circle, then as shown in FIG. 4 'the edge information formation image generated by the edge information formation mechanism 15 is input to the object / background separation mechanism 15 16. Input the object area image 142 generated by the object / background separation mechanism 16 into the position / size detection mechanism 17 'Based on this', the position / size detection mechanism 17 can calculate and display the area of the moving object 2 and the stationary object 3 of a specific size The number of pixels at the position. In other words, the 'all object counting section 02' can output the number of all objects with higher accuracy. In addition, as described above, the visual device can calculate a moving image by using the moving object counting unit 101 that counts the number of moving objects 2 in the frame image 1 of the moving image, and all the object counting units 102 having approximately the same mechanism. The total number of moving objects 2 and stationary objects 3 in 1. Therefore, a method for simultaneously calculating the moving object 2 and the stationary object 3 using one visual device is considered. At this time, if the number of the moving objects 2 and the stationary objects 3 can be obtained, the ratio of the moving objects 2 or the stationary objects 3 can be easily calculated for all the objects. Hereinafter, a visual device for determining the ratio between the moving object 2 and the stationary object 3 will also be described. The overall configuration of the visual device that calculates the number of moving objects 2 and all objects and the ratio of moving objects 2 and stationary objects 3 is shown in FIG. 5 'It is roughly divided into a moving object counting section 101 and all objects The counting unit 102 is composed. Moving Object Counting Unit 101 and All Object Counting Units 02, 45 ---------------------- Order ----------- line (please Please read the notes on the back before filling this page.) This paper size is applicable to China National Standard (CNS) A4 (21〇X 297 meals) 540010 A7 B7____ 5. Description of the invention (¥ () (Please read the note on the back first (Please fill in this page again) is composed of each of the aforementioned mechanisms. “The image acquisition mechanism 11 and image memory mechanism 12 of the moving object counting unit 101 and all the object counting units 102 perform exactly the same operation”. In FIG. 5, the common image acquisition mechanism 11 and the image counting mechanism 12 ° are shared. Therefore, “the important thing is that the pixel count holding mechanism 19 of the moving object counting unit 101” is input from the pixel counting mechanism 18 of all the object counting units 102. Number of all objects 'The pixel number holding mechanism 19 of the all object counting section 102 is the number of moving objects input from the pixel counting mechanism 18 of the moving object counting section 101.' Using external signals such as ratio switching signals, it calculates moving objects 2 and stationary objects. 3 scales to switch to float small Counting points, etc. Of course, 'if it is not necessary to calculate the ratio of the moving object 2 and the stationary object 3' or can be realized by other devices, it is not necessary to add this function to the pixel number holding mechanism 19. On the contrary, when the In the case of the number of moving objects and the number of stationary objects, it is only necessary to add corresponding functions to the pixel number holding mechanism 19. By using a general processor in the pixel number holding mechanism 19, the visual device can freely make such changes depending on the application. Printed by the Ministry of Intellectual Property Bureau's Consumer Cooperatives and 'as explained above' from the application of 靑 patent 轫 轫 1 to the scope of patent application of the visual device (the moving object counting section 101 and all object counting section 102 . It is a device used to calculate the number of moving objects 2 and stationary objects 3 in the block diagram image 1 taken by the camera. Therefore, if the moving object 2 and the stationary object 3 are included in the frame image 1, these visual devices can calculate the number of the moving object 2 and the stationary object 3. However, if the moving object 2 and the stationary object 3 deviate from the frame image 1, these visual devices cannot calculate the number of the moving object 2 and the stationary object 3. Therefore, in the following, the mobile camera 10 (refer to Figure 6) is used, and 46 ^ paper size is always applicable to China National Standard (CNS) A4 specifications ¢ 210 X 297 mm 540010 Α7 Β7 V. Description of the invention (Search for moving objects 2 and stationary objects 3 at t0, while calculating the number of patent applications ranging from patent application scope item 4 to patent application scope item 5. First, mobile camera 10 There are mechanisms for moving in the horizontal and vertical directions. By inputting instructions for controlling the movement angle from the outside, each lens can be panned and tilted. The mobile camera 10 is also equipped with a mechanism for changing the magnification of the captured image. The external zoom control command enables zooming. Therefore, the mobile camera 10 can finely vibrate the camera according to the external command. Accordingly, the frame image 1 of the moving image captured by the mobile camera 10 Shake occurs, and the object in the frame image 1 is captured as if it were vibration. That is, if the mobile camera 10 is used That is, it is not necessary to use the image vibration mechanism 13 of the all-object counting unit 102, and only the moving object counting unit 101 can calculate the area of all objects. This method uses the physical mechanism of the mobile camera 10 in processing speed and counting. Although there are problems in accuracy, etc., since only the moving object counting unit 101 can calculate the area of both the moving object 2 and all objects, it is suitable for calculating the moving objects 2 and 2 located in a spacious place that cannot be included in the frame image 1. Stationary object 3. Here, 'moving camera 10 can output the current position moved according to movement instructions such as panning, tilting, and zooming, and output as needed, and can also move or stop now if necessary The person who outputs the status of the mobile camera 10. At this time, if the vision device can control the pan, tilt, and zoom of the mobile camera 10, 'the vision device can change the direction and magnification of the mobile camera 10' to make the mobile Camera 1047 This paper size is applicable to China National Standard (CNS) A4 specification v (210 X 297 public love) ------ · -------------- --------- Line (please read the precautions on the back before filling this page) 540010 Α7 Β7 V. Description of the invention It is always possible to shoot moving objects 2 and stationary objects 3 with appropriate sizes. Therefore, Figure 6 The non-movable video camera 10 is equipped with a visual device of the scope of patent application No. 4 of a basic mechanism that can constantly photograph a moving object 2 and a stationary object 3 with an appropriate size. At this time, the visual device is used according to its function. Three coordinate systems, that is, the camera coordinate system, the image coordinate system, and the environmental coordinate system. First, the camera coordinate system, as its name implies, is used by mobile cameras to control the moving lens, tilt and zoom with each minimum control unit. Three-dimensional ball coordinate system inside the camera. Generally speaking, the origin of a camera coordinate system is the inherent position of a mobile camera called the home position. The camera coordinate system is the only coordinate system capable of displaying the physical positions of the moving object 2 and the stationary object 3. Even so, the camera coordinate system is not general because it is different in each type of mobile camera 10. Therefore, the vision device must conceal the camera coordinate system from other mechanisms. Second, the image coordinate system is a two-dimensional coordinate system in pixels using the center of the frame image 1 captured by the mobile camera 10 as the origin. This is used to show whether the moving object 2 and the stationary object 3 are located on the pixel in the frame image 1. Therefore, although the image coordinate system is suitable for distinguishing the fine positions of plural objects located in the frame image 1, the physical position of the moving object 2 and the stationary object 3 cannot be displayed only by the image coordinate system. Third, the environmental coordinate system is a three-dimensional ball coordinate system used by the visual device to uniformly display the positions of the moving object 2 and the stationary object 3 internally. The environmental coordinate system uses angles in radians in the horizontal and vertical directions. In order to display the product of the size of the object and the distance from the object, a distance of 1. 0 is 48 This paper size applies to China National Standard (CNS) A4 specifications 210 x 297 mm) (Please read the precautions on the back before filling this page) > 衣 -------- Order ---- ----- Line "Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 510010 Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7 V. Invention Description (ep unit real number. Generally speaking, due to the size of the object will not Extreme changes, so the distance to the object can be regarded as proportional to the magnification of the mobile camera 10. The origin of the environmental coordinate system is arbitrary. That is, in principle, the environmental coordinate system is used to display any arbitrary on the environmental coordinate system Relative coordinates of 2 points. The visual device uses a mobile camera 10 to distinguish multiple objects by projecting objects in a photographable environment into an environmental coordinate system. Therefore, the camera coordinate system and image coordinate system must be different from each environment. The coordinate system converts each other. The institutions that assume this role are the camera / environmental coordinate conversion mechanism 20, the image / environmental coordinate conversion mechanism 21, and the motion control mechanism 23. These institutions The specifications of the self-moving camera 10 and the image acquisition mechanism 11 are used to obtain the camera coordinate system and the image coordinate system to calculate the matrix for conversion to the environmental coordinate system. Alternatively, the self-camera coordinate system can also be calculated The inverse matrix of the conversion matrix to the environmental coordinate system is used to find the conversion matrix from the environmental coordinate system to the camera coordinate system. However, since the origin of the camera coordinate system is the origin position of the mobile camera 10, it is converted from the camera coordinate system The position of the environmental coordinate system is the relative position of the origin position of the mobile camera 10 on the environmental coordinate system. On the other hand, the position of the camera coordinate system converted from the environmental coordinate system is the position of the camera coordinate system. The relative position of the current position of the mobile camera 10. Furthermore, since the image coordinate system is a two-dimensional coordinate system, it cannot be converted into an environmental coordinate system using only the image coordinate system. Therefore, the image / environment coordinate conversion mechanism 21, Is the direction and direction of the mobile camera 10 displayed using the environmental coordinate system Rate, and the size of the moving object 2 and the stationary object 3 in the frame image 1, can be calculated at any time with 49 paper standards applicable to China National Standard (CNS) A4 specifications (210 X 297 mm) * ------- ---------- Order --------- line (please read the precautions on the back before filling this page) 540010 Printed by A7 B7, Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Finance (/), The conversion matrix is converted from the image coordinate system to the environmental coordinate system. In addition, the position of the environmental coordinate system converted from the image coordinate system becomes the relative position from the image center. Therefore, a patent is applied for The other mechanisms of the visual device in the scope of item 4 can be regarded as for the purpose of constantly photographing the moving object 2 and the stationary object 3 with appropriate sizes, and used to generate and convert the positions of the moving object 2 and the stationary object 3 displayed by the three coordinate systems mechanism. Therefore, based on the camera coordinate system, image coordinate system, and environmental coordinate system of the coordinate system, FIG. 6 will be described. First, the image acquisition mechanism 11, the edge information generation mechanism 14, the position / size detection mechanism 14, and the size detection mechanism π are related to the moving object counting section 101 and the all object counting section 102. The three items described in the visual device are the same. However, in the visual device of the scope of application for items 1 to 3, the formation edge information image U5 generated by the generation of the edge information formation mechanism 15 is input to the position / size detection mechanism 17; In the visual device, the coarse edge information_image 113 generated by the edge information generation mechanism 14 is input to the position / size detection mechanism 17. Of course, in this visual device, although the edge information forming mechanism 5 can be used to form the edge information #image 115 into the fii setting / size detection mechanism 17, the performance of the visual device is as follows, Even if the edge information forming mechanism 15 is not used, it is enough to be the first. In this visual device, it is not necessary to calculate the number of fields of the moving object 2 and the stationary object 3 like the visual device of the scope of application patent items 1-3. On the contrary, it is more important in this visual device to search for a moving object 2 1 円 continue object 3, and align the direction and magnification of the mobile camera 10 with the words of the evening 50. This paper standard applies to China National Standard (CNS) A4 Specifications (210 X 297 g) --- ^ ---------- install -------- order --------- line (Please read the precautions on the back before filling in this Page) 540010 A7 B7 5. Direction of the invention (4). Moreover, the position / size detection mechanism 17 can also obtain the approximate size of the number of areas of the moving object 2 and the stationary object 3 from the coarse edge information 112. Second, The edge information forming mechanism 15 uses the rough edge information image 1Π generated by the edge information generating mechanism 14 to generate the edge information image 115. That is, while the edge information forming mechanism 15 generates the edge information image 115, The moving object 2 will move. Generally speaking, since the mobile camera 10 has a physical mechanism, the moving speed of the mobile camera 10 will not be too fast. Therefore, when the visual device moves the mobile camera 10, Object 2 moves at a speed that is too late Possibility of mobile camera 10. Due to the above reasons, the vision device for patent application No. 4 does not necessarily require clear and clear formation of edge information 114 for moving objects 2 and stationary objects 3. Next, the camera / environment coordinate conversion mechanism 20 and The image / environment coordinate conversion mechanism 21 converts the position of the moving object 2 and the stationary object 3 displayed by the repeated information image 132 generated by the position / size detection mechanism 17 into the position of the environmental coordinate system. At this time, If the total area position of the moving object 2 and the stationary object 3 in the frame image 1 is 2 or more, there are also more than two positions on the environmental coordinate system. Therefore, in order to control the moving lens, tilt and zoom of the mobile camera 10 In any of these, an object is aligned with the direction and magnification of the mobile camera 10, and one of the positions on the environmental coordinate system must be selected. The position selecting mechanism 22 selects a position on the environmental coordinate system according to a certain judgment criterion. The judgment criteria used here are mainly as follows. First, select the nearest (or most recent) object on the environmental coordinate system. Large). This is because when the object is far away (or smaller), the edge information is 51. This paper size is applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) -------- · ------ ^ Clothing (Please read the precautions on the back before filling out this page) Order --------- Line · Printed clothing by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 510010 Intellectual Property of the Ministry of Economic Affairs Printed by the Consumer Cooperative of the Bureau A7 B7 V. Description of the invention (h) Noise may be generated by the establishment 14, so try to choose the one with the highest probability of having objects on the environmental coordinate system. Second, when the environmental coordinate system When a plurality of positions are concentrated within a certain range, select one of them. There are two possibilities. One is the possibility that the edge information generating mechanism 14 generates scattered coarse edge information 112 for one object, and the other is the possibility that there are actually multiple objects. Third, when most positions on the environmental coordinate system are approximately the same distance (or approximately the same size), the position closest to the origin is selected, that is, the position closest to the direction of the mobile camera 10 is selected. Depending on the use and situation, the above-mentioned matters are appropriately combined, and the position selecting mechanism 22 can select a position on the environmental coordinate system. In addition, the visual device of the fourth scope of the patent application, in addition to the position selected by the position selection mechanism 22, can also obtain a pan, tilt, and zoom mobile camera 10 at the following positions. First, the vision device must cause the mobile camera 10 to vibrate. Therefore, in order to vibrate the mobile camera 10, the vibration command generating mechanism 25 specifies the position where the mobile camera 10 moves as the position on the environmental coordinate system. The position designated by the vibration command generating mechanism 25 is a range in which the mobile camera 10 does not vibrate, and is determined by a pseudo-random number or the like. Furthermore, the visual device in the fourth scope of the patent application is obtained by inputting a control command from the outside to obtain a pan, tilt and zoom of the mobile camera 10. As for general use, since the moving camera 10 is currently facing the direction and magnification to move the lens, tilt, and zoom, using the control command input mechanism 24 to temporarily store the control instructions, the current position of the mobile camera 10 is calculated. The position on the environmental coordinate system as the origin. In addition, by improving the control instruction input machine 52, this paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm) --- ^ ------------------ Order --------- line (please read the notes on the back before filling this page) 540010 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7 V. Description of the invention ((r) Structure 24, can be easily To move the mobile camera 10 to a specific position. Therefore, in order to control the moving lens, tilt, and zoom of the mobile camera 10, the motion control mechanism 23 selects one of three positions on the aforementioned environmental coordinate system. The motion control mechanism 23 When selecting a position from the three positions on the aforementioned environmental coordinate system, the order of control command input mechanism 24, position selection mechanism 22, and vibration command generation mechanism 25 is selected by the mechanism having its own input position. The selected position Converted from the environmental coordinate system to the camera coordinate system. After that, the camera command generating mechanism 26 is used to convert the commands recognized by the mobile camera 10 to the mobile camera 10. According to this, the scope of patent application The four visual devices can control the moving lens, tilt and zoom of the mobile camera 10. Depending on the model of the mobile camera 10, the mobile camera 1 cannot be obtained without sending a specific instruction to the mobile camera 10. 〇 State ', that is, information on the direction and magnification of the mobile camera 10 and whether the mobile camera 10 is moving or stopped cannot be obtained. Therefore, the motion control mechanism 23 selects one of three positions and outputs it to After the camera instruction generation mechanism 26, instruct the camera instruction generation mechanism 26 to transmit an information instruction asking whether the mobile camera 10 is moving to the mobile camera 10 'from the mobile camera 10 through the camera / environment coordinate conversion mechanism 20' Stand by until the information is received. If the received information is displayed while moving, that is, once again instruct the camera command generation mechanism 26 to transmit the information command asking whether the mobile camera 10 is moving to the right of the mobile camera 10 The date of the fund only shows that the camera is not stopped, that is, the camera command generating mechanism 26, Will inquire about the current position of the mobile camera 10 and 53 of this magnification. This paper size is applicable to the Chinese National Standard (CNS) A4 specification coffee 297 public love. —1 — · -------------- Order- -------- Line (Please read the notes on the back before filling this page) 540010 Printed by A7 B7, Consumer Cooperative of Intellectual Property Bureau, Ministry of Economic Affairs V. Description of invention (0) The instruction is transmitted to mobile camera 10. During this period, the motion control mechanism 23 does not select three positions. The camera instruction generating mechanism 26 transmits a corresponding instruction to the mobile camera 10 according to an instruction from the motion control mechanism 23. The camera / environmental coordinate conversion mechanism 20 transmits the information of whether the mobile camera 10 is moving to the motion control mechanism 23 intact, and converts the current position and magnification of the mobile camera 10 from the camera coordinate system to the environmental coordinate system. position. According to this, the visual device of the fourth scope of the patent application can inspect the state of the mobile camera 10 one by one while controlling the moving lens, tilt and zoom of the mobile camera 10. So far, a visual device has been described which includes a mobile camera 10 that constantly captures a moving object 2 and a stationary object 3 in a frame image 1 in an appropriate size in an appropriate size. However, it does not mean that the moving object 2 and the stationary object 3 often exist in the range captured by the mobile camera 10, and the moving object 2 will soon move from the range captured by the mobile camera 10 to another position. Of course, if the moving object 2 moves, the shape and color of the moving object 2 seen from the moving camera 10 will of course change. Therefore, the visual device in the scope of patent application No. 5 described below can be regarded as the visual device in the scope of patent application No. 4, which is added to more accurately identify the moving object 2 and the stationary object 3 in the frame image 1. Of several shapes and colors, and several mechanisms for more accurately measuring the number of moving objects 2 and stationary objects 3 taken by the mobile camera 10 in a fixed time, and more accurately making the mobile camera 10 is a mechanism that moves toward the position of the moving object 2 and the stationary object 3 within the range that the moving camera 10 can take. 54 This paper size applies to China National Standard (CNS) A4 specification C210 X 297 mm)-^ I L. ---. ------ ^^ 衣 -------- Order --------- line 2 Qing first read the phonetic on the back? Please re-fill this page on the matter} 540010 Printed by A7 B7, Consumer Cooperatives, Intellectual Property Bureau, Ministry of Economic Affairs V. Description of Invention (fj) The visual device in Figure 7 is an edge information formation mechanism added to the visual device in the fourth patent application scope. 15. Object / background separation mechanism 16, domain normalization mechanism 27, and normalized image holding mechanism 28 to generate a normalized image 145. Edge information formation mechanism 15, object / background separation mechanism 16, and edge information formation mechanism 15 of the visual device of the patent application scopes 1 to 3 composed of the moving object section 101 and all object counting sections 102, The object / background separation mechanism 16 is the same. When the domain normalization mechanism 27 inputs the object domain image 142 and the digital image 111 from the object / background separation mechanism 16 and the image acquisition mechanism 11, respectively, the separated object domain 143 is cut out from the digital image 111 to separate them. The object area 143 is deformed, and the normalized image 145 is generated using interpolation and enlargement as much as possible while matching the image size of the digital image 111. Since the domain normalization mechanism 27 processes only the vicinity of each pixel, it can normalize the separated object domain 143, so it is suitable for parallelization. If the domain normalization mechanism 27 can output all pixels of the normalized image 145 in parallel, the communication from the domain normalization mechanism 27 to the normalized image holding mechanism 28 can perform each pixel in parallel. When the normalized image holding mechanism 28 inputs the normalized image 145 from the domain normalization mechanism 27, if the output end of the normalized image 145 requires the normalized image 145 in an appropriate format, the normalized image 145 can be Convert to the format required by the output of the normalized image 145. After that, the normalized image holding mechanism 28 memorizes the normalized image 145 for a predetermined period before the normalized image 145 is surely transmitted to the output end of the normalized image 145. If the normalized image holding mechanism 28 restricts the conversion format, since only 55 pixels are used, this paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm) n —wn I 11 «» 11 n ϋ ϋ I * n · ϋ n (Please read the notes on the back before filling out this page) Order --------- line- 540010 Printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7 V. Invention Description (α) Nearby processing can convert the normalized image 145, so it is suitable for parallelization. If the normalized image holding mechanism 28 can output all pixels of the normalized image 145 in parallel, communication from the normalized image holding mechanism 28 to the output terminal of the normalized image 145 can be performed in parallel with each pixel. Therefore, by using the edge information forming mechanism 15 'object / background separation mechanism 16, domain normalization 27, and normalized image holding mechanism 28, even if the moving object 2 and the stationary object 3 are captured by the visual device of the patent application item 4 In the frame image 1, an arbitrary position exists in an arbitrary size, and the visual device of FIG. 7 can also generate a normalized image 145 of the moving object 2 and the stationary object 3 as much as possible. Accordingly, when the output end of the normalized image 145 must recognize the moving object 2 and the stationary object 3, the output end of the normalized image 145 does not need to consider the positions of the moving object 2 and the still object 3 in the frame image 1. And size, and appropriate identification methods can be used. In the visual device shown in FIG. 7, among the edge information generating means 14, edge information forming means 15, object / background separation means 16, domain normalization 27, and normalized image holding means 28, the images inputted by each mechanism Image resolution or image size does not have to be the same. For example, in the visual device of FIG. 7, a low-resolution digital image 116 that reduces the resolution of the digital image 111 is input into the edge information generating mechanism 14. On the other hand, in the edge information forming mechanism 15, an appropriate method is input. The image size of the low-resolution coarse edge information image 117 generated by the edge information generation mechanism 14 is enlarged to the coarse edge information image 113 of the image size of the digital image 111 in the object / background separation mechanism 16 and the field standardization mechanism. By inputting a digital image 111 in 27, the load on the edge information generating mechanism 14 can be reduced. That is, almost 56 paper sizes are applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) ------ 1 ------ installation -------- order --- ------ line (Please read the notes on the back before filling this page) Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 540010 A7 B7 V. Description of the invention The quality of the normalized image 145 can control the moving lens, tilt, and zoom of the mobile camera 10 after the position / size detection mechanism 17 at a higher speed. If the method is further performed, the edge information forming mechanism 15 In the input rough information information image 117 generated by the input edge information generation mechanism 14, the rough information information image 119 is cut out from the area where the rough edge information 112 exists, and the object / background separation mechanism 16 and the field specifications In formula 27, inputting a digital image 120 cut out from the digital image 111 in the same position area as the rough edge information image 119, that is, reducing the load generated by the normalized image 145 of the edge information forming mechanism 15 after . Here, if the scope of patent application can be used The four-vision vision device enables the mobile camera 10 to capture moving objects 2 and stationary objects 3 at an appropriate size in the center of the frame image 1 to determine the rough edge information image 119 of the digital image 111 in advance. And the cut-out area for cutting out the digital image 120. Accordingly, the visual device of FIG. 7 can use one mobile camera 10 to achieve the same performance as an object search device using a wide-angle camera and a high-resolution camera. The vision device of FIG. 8 is a person who adds an image recognition mechanism 29 and a recognition mechanism holding mechanism 30 to the vision device of FIG. 7 to generate a recognition result. When the image recognition mechanism 29 inputs a normalized image from the domain normalization mechanism At 145, an appropriate image recognition method is used to identify the specification area 144 of the moving object 2 and the stationary object 3 of the normalized image 145, and the recognition result is output. Since the normalized image 145 of the input image recognition mechanism 29 is used, The field standardization mechanism 27 deforms the moving object 2 and the stationary object 3, because 57 paper sizes apply the Chinese National Standard (CNS) A4 specification (210 X 297 mm) — III — 1 · 1 IIII · 1111111 «— — — — — — II Αν (Please read the precautions on the back before filling out this page) 540010 A7 B7______ V. Description of the invention Lrt) (Please read the precautions on the back before filling out this page) This picture Instead of using a stroke extraction method, a Fourier transform, a Harvard transform, and the like to perform a geometric analysis of the better position deviation, the image recognition mechanism 29 is better than an image matching method that compares an input image with a template image. As an image matching method, a neutral network capable of learning the propagation of a template image or the like by using an error back propagation method can be used. In addition, when using a neutral network, it is also possible to perform parallelization and speedup by using an accelerator of the neutral network. The recognition result 'of the normalized image 145 is output from the image recognition mechanism 29 to the recognition result holding mechanism 30. When the recognition result holding mechanism 30 inputs the recognition result of the normalized image 145 from the image recognition mechanism 29, if the output of the recognition result requires a signal of an appropriate format, the recognition result is converted into the format required by the recognition result output. After that, the recognition result holding mechanism 30 records the recognition result for a certain period before the recognition result can be surely transmitted to the recognition result output terminal. Therefore, by using the image recognition mechanism 29 and the recognition result holding mechanism 30, the vision device of FIG. 8 can generate the recognition results of the moving object 2 and the stationary object 3 captured by the mobile camera 10 at an appropriate size. According to this, the output end of the recognition result can use the vision device of FIG. 8 as the recognition device of the moving object 2 and the stationary object 3 captured by the mobile camera 10. The visual device in Figure 9 is printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. The visual device in Figure 8 is to increase the environmental understanding mechanism 31, the timing mechanism 32, and the environmental map agency 33 to generate an environmental map. The timing mechanism 32 uses a timer circuit to output the time in units of 1 centimeter (mm) seconds. At this time, the time is continuously output from the timing mechanism 32 to the environmental understanding mechanism 31. 58 'This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 public love) — " " 540010 A7 B7 V. Description of the invention (q) When the environment understanding mechanism 31 inputs the moving object from the image recognition mechanism 29 When the recognition results of 2 and the stationary object 3 are obtained, environmental data composed of the recognition result, the position of the mobile camera 10 on the environmental coordinate system, and the current time is created. In addition, when the environment understanding mechanism 31 converts the image / environmental coordinate conversion mechanism 21 and the fields of all the moving objects 2 and stationary objects 3 in the input frame image 1 on the environmental coordinate system, it only creates the moving objects 2 and Environmental data of the number of fields of the stationary object 3, the environmental data is the recognition result composed of invalid data, the position of the mobile camera 10 on the environmental coordinate system plus one of the moving object 2 and the stationary object in the frame image 1 The position of the field of 3 on the environmental coordinate system, and the current time. At this time, the environmental map is a collection of environmental data made a certain time before the present time, and the position in the environmental data is represented by an environmental coordinate system using the origin of the mobile camera 10 as the origin. The environmental understanding mechanism 31 ′ is to add and delete environmental data to and from the environmental map over time. In addition, the environmental understanding mechanism 31 deletes the duplicated environmental data, and when the position in the environmental data whose identification result is invalid data is near the position in other environmental data whose identification result is not invalid data, the identification result is deleted as Environmental data for invalid data. Furthermore, when the position in the environmental data whose identification result is not invalid data is near the position in other environmental data where the identification result is not invalid data, if the identification results in the environmental data are consistent, the former environment is deleted. Data. The accuracy of the environmental map is determined based on the recording time of the environmental data and the range near the location in the environmental data. The environment map is output from the environment understanding means 31 to the environment map holding means 33. 59 This paper size is in accordance with Chinese National Standard (CNS) A4 (21〇x 297 mm) --------------, cans (Please read the precautions on the back before filling this page ) ---- Order --------- Line · Printed by the Consumers' Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs and printed on the consumer cooperation of the Intellectual Property Bureau of the Ministry of Economic Affairs Du printed clothing 510010 A7 B7 V. Description of the invention (㈡?) When the environment map holding mechanism 33 inputs the environment map from the environment understanding mechanism 31, if the output terminal of the environment map requires a signal of an appropriate format, the environment map is converted into the format required by the output terminal of the environment map. Then, the environment map holding mechanism 33 memorizes the environment map for a certain period of time before the environment map is surely transmitted to the output end of the environment map. Therefore, by using the environment understanding mechanism 31, the timing mechanism 32, the environment map holding mechanism 33, and the visual device of FIG. 9, the positions of the moving object 2 and the stationary object 3 captured by the mobile camera 10 can be specified. According to this, the output terminal of the environment map can use the visual device of FIG. 9 as a specific device for the positions of the moving object 2 and the stationary object 3 which can be captured by the mobile camera 10. The vision device of FIG. 10 is an object position estimation mechanism added to the vision device of FIG. 9 to control the mobile camera 10 at the estimated position of the object. When the object position estimation mechanism 34 inputs an environment map from the environment understanding mechanism 31, it selects an environmental data whose identification result is invalid data, and extracts the position in the environmental data. From this position, the current position on the environmental coordinate system of the mobile camera 10 calculated by the camera / environment coordinate conversion mechanism 20 is subtracted from the object position estimation mechanism 34, so that the environmental coordinate system of the mobile camera 10 can be obtained. The current position is used as the origin. Even though the rough edge information 112 has been generated in the past, the recognition result is the relative position of the moving object 2 and the stationary object 3 that have not appeared on the environmental coordinate system. However, in the case of the moving object 2, the moving object 2 does not necessarily exist at the current position. Therefore, it is possible that there are moving objects 2 and stationary objects at this position. This paper size is applicable to China National Standard (CNS) A4 (210 X 297 mm) --- Γ --- * ------ ^^ Install -------- order --- I ----- line (please read the precautions on the back before filling this page) 540010 A7 B7 V. Description of the invention (4) Moving object 2 of body 3 and The estimated position of the stationary object 3. On the other hand, if there is no environmental data in the environmental map whose identification result is invalid data, that is, a position on the system of an appropriate environmental coordinate system is generated within a range where the mobile camera 10 can move. In this case, as a basis for generating an appropriate position, the evaluation rate is as follows. First, use pseudo-random numbers to generate arbitrary positions. Second, calculate low-density locations in the environmental data in the environmental map. Third, the positions within the range where the mobile camera 10 can move are sequentially generated in an appropriate order. For example, within the movable range of the mobile camera 10, positions are generated in order from the upper left position to the right, and then descend one segment after reaching the right end, and then sequentially generate positions to the left, descend one segment after reaching the left end, and then, The right generates positions in sequence, repeating this action. Depending on the application and the situation, these standards can be combined to effectively estimate the estimated positions of the moving body 2 and the stationary object 3 that were not captured by the mobile camera 10. The object estimation position is output from the object position estimation mechanism 34 and output to the motion control mechanism 23. Here, the motion control mechanism 23 is based on the motion control mechanism 23 of the visual device of the scope of application for patent, and is modified as follows. That is, in order to control the moving lens, tilt, and zoom of the mobile camera 10, the motion control mechanism 23 uses the order of the control instruction input mechanism 24, the object position estimation mechanism 34, the position selection mechanism 22, and the vibration instruction generation mechanism 25. Select the mechanism of the input location. However, the motion control mechanism 23 must discontinuously select the input position from the object position estimation mechanism 34. Therefore, by using the object position estimation mechanism 34 and the visual device of Fig. 10, it is possible to search for the moving object 2 and the stationary object 3 existing in the shooting range of the mobile camera 10. According to this, the output end of the environmental map can use 61 paper sizes to comply with China National Standard (CNS) A4 (210 x 297 mm) ------- J ------ (Please read the back Please note this page before filling out this page) Order ------ Printed by the Consumers' Cooperatives of the Intellectual Property Bureau of the Ministry of Economy 540010 A7 B7 Device. As a specific device for the positions of the moving animal body 2 and the stationary object 3 which can be captured by the mobile camera 10. The visual device of FIG. 11 is an object counting mechanism 35 and an object number holding mechanism 36 added to the visual device of FIG. 10 to generate an arbitrary number of objects. When an object § ten organization 35 inputs an environment map from the environment understanding organization 31, it can calculate the number of environmental data with the recognition result of the meaning of a specific object in the environment map to generate an arbitrary number of objects. As for the recognition result in the environmental data, an arbitrary number can be selected from the types of objects recognized by the image recognition mechanism 29. The arbitrary number of objects is output from the object counting mechanism 35 to the object number holding mechanism 36. If necessary depending on the application, by changing the object counting mechanism 35, it is possible to easily specify the types of objects to be counted from the outside. When the number-of-objects holding mechanism 36 inputs an arbitrary number of objects from the number-of-objects mechanism 35 ', before the arbitrary number of objects is surely transmitted to the output terminal of the arbitrary number of objects, it records a certain amount of time. Therefore, by using the visual device of the object counting mechanism 35 and the number-of-objects holding mechanism 36 'FIG. Π, it is possible to count the number of the specific moving objects 2 and the stationary objects 3 existing in the shooting range of the mobile camera 10. According to this, the arbitrary object number output terminal can use the vision device of FIG. 11 as a specific device for the positions of the moving object 2 and the stationary object 3 which can be captured by the mobile camera 10. As shown in Fig. 12, the visual device in the scope of the patent application No. 5 is a geometric analysis mechanism 37 added to the visual device of Fig. 11 to accurately and quickly create an environment map. 62 This paper size applies to China National Standard (CNS) A4 (210 X 297 mm) (Please read the precautions on the back before filling out this page)-Line-540010 Printed by the Consumers ’Cooperative of Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7 V. Explanation of the invention (to) When the geometric analysis mechanism 37 inputs the edge information image 115 from the edge information formation mechanism 15 to calculate the geometric analysis in the frame image 1 by performing geometric analysis such as the stroke extraction method, Fourier transformation, and Harvard transformation. The shapes of the moving object 2 and the stationary object 3 to generate a geometric analysis result. In the past image recognition, the line part was directly extracted from the image itself, or various transformations that did not affect the position and size were used to identify the objects in the image. Therefore, although the processing is very simple, the number of combinations is huge, and the result requires a huge amount of calculation. It takes a long time before the desired result is obtained, or only low-quality identification results can be obtained. However, the visual device in the fifth item of the patent application is a graphic recognition of the moving object 2 and the stationary object 3 in the frame image 1. The object / background separation mechanism 16, the domain normalization mechanism 27, and the image recognition mechanism 29 Go ahead. Therefore, the geometric analysis mechanism 37 uses graphics such as the edge information image 115, the near-object / background separation mechanism 16, the domain normalization mechanism 27, and the image recognition mechanism 29 which are difficult to cope with. The analysis can be roughly performed by a simple method, that is, unnecessary image matching can be omitted by the image recognition mechanism 29, and an environment map can be accurately and quickly created by the environment understanding mechanism 31, and the load of the geometric analysis mechanism 37 itself can be reduced. The geometric analysis result is output from the geometric analysis mechanism 37 to the image recognition mechanism 29 and the environment understanding mechanism 31. Therefore, the image recognition mechanism 29 and the environment understanding mechanism 31 respectively perform the following operations. When the image recognition mechanism 29 inputs a geometric analysis result from the geometric analysis mechanism 37 ', first, it is determined whether it is a figure of an object recognized by the image recognition mechanism 29. If the geometric analysis result is an out-of-object figure, the image recognition machine 63 paper size is applicable to China National Standard (CNS) A4 specification (210 X 297 mm) ------------ install --- ----- Order -------- line (please read the precautions on the back before filling this page) 540010 A7 B7 V. Description of Invention (tl) Structure 29 will not operate. If the result of the geometric analysis is the object figure, the image recognition mechanism 29 uses the template image for the object figure to perform the pattern matching. For example, 'perxeptron for learning by error back propagation method, learn the visual controller for each object figure to create learning materials, and then use the geometric analysis results to select learning materials' for the object The graphics are effectively pattern matched. That is, if the result of the geometric analysis is a circle, that is, only the learning materials of circular objects such as faces and balls are used. If the result of the geometric analysis is a triangle, that is, only the learning materials of railway signs and triangular objects such as mountains are used. In this way, not only can a small visual controller recognize patterns of most objects, but the visual controller itself can also learn effectively. When the environment understanding mechanism 31 inputs the geometric analysis result from the geometric analysis mechanism 37, first, it is determined whether it is a figure of an object recognized by the image recognition mechanism 29. If the geometric analysis result is an out-of-object figure, the environmental understanding mechanism 31 immediately deletes the environmental data from the environmental map (located near the position on the environmental coordinate system of the mobile camera 10, and the identification result is environmental funds with invalid data). Accordingly, unnecessary environmental data can be deleted from the environment map, even if unnecessary object estimation positions such as the object position estimation mechanism 34 are not output. When the result of the geometric analysis is the object figure, it stands by until the recognition result is input from the image recognition mechanism 29. Therefore, by using the geometrical analysis mechanism 37 and the vision device of the fifth patent range, the number of specific moving objects 2 and stationary objects 3 existing in the range that the mobile camera 10 can capture can be accurately and quickly calculated. According to this, the output terminal of any number of objects can use 64 in the scope of patent application No. 5 -------------- (Please read the precautions on the back before filling this page) Order- -------- Printed by the Consumers' Cooperatives of the Bureau of Intellectual Property, Ministry of Economic Affairs, Paper Size Applicable to Chinese National Standard (CNS) A4 (210 x 297 mm) 540010 A7 B7 V. Description of Invention (M) Visual Device As a high-speed device for calculating the number of specific moving objects 2 and stationary objects 3 that the mobile camera 10 can capture. (Please read the precautions on the back before filling out this page.) In addition, for the visual devices from the first to the fifth of the patent application scope, the image memory mechanism 12, image vibration mechanism 13, and edge information generation mechanism 14 are used. The edge information forming mechanism 15, the object / background separation mechanism 16, the position / size detection mechanism Π, the domain normalization mechanism 27, the normalized image holding mechanism 28, and the image recognition mechanism 29 can be used by the array operation unit 40 (Array Operation Unit), and assemble it. Therefore, in the following, an embodiment of the data processing device 110 using the array operation unit 40 is taken as an example, and the visual device from the 6th to the 12th patent application scope will be described with reference to the drawings. Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. First, the array operation unit 40 uses one pixel of the input image and its nearby pixels to generate one pixel of the output image. Therefore, as shown in Fig. 13, the array operation unit 40 is adapted to the size of the input image, and the data processing device 110 is arranged in a grid pattern. The data processing device 110 can generate an output image from the input image. In Fig. 13, the array operation unit 40 is abbreviated as A0U. Next, the array operation unit 40 may be assembled using dedicated hardware, or may be assembled using software on a general computer. That is, if an output image can be generated from an input image, the assembly mechanism is not limited. Therefore, by the algorithm of the display array operation unit 40, the image processing of the data processing device 110 can be displayed. Therefore, in order to display the algorithm of the array operation unit, the image memory mechanism 12 (refer to Figs. 1 and 6), the image vibration mechanism 13 (refer to Figs. 3, 4 and 5), and edge information are generated. Institution 14 (refer to Figures 1 and 6), edge information formation mechanism ι5 (refer to 65 moderate ruled paper I format 4) AS) N (C standard 540010 Α7 Β7 V. Description of the invention) (Please read the back first Please fill in this page for the precautions) (see Figures 1 to 5 and 7 to 12), object / background separation mechanism 16 (refer to Figures 2, 4, 5, 7 to 12), position / size detection mechanism 17 ( (See Figures 1 and 6), the domain normalization mechanism 27 (see Figure 7), the normalized image holding mechanism 28 (see Figure 7), and the image recognition mechanism 29 (see Figure 8)-used To explain. Let any 2n hue of width w, height h, and number of bands b be X, y, w. X, y, and W are band pixels 値 Xuk, yuk, wuk using each position p (i, j, k), as shown in the formula 1, Equation 2, Equation 3. Also, bold text is the display vector. In addition, 11 is a non-negative integer, and ~,} 1, 13, Bu ",] ^ are natural numbers. X = is value at k), l < i < w, l < j < h, l < k < b] (1) y = (Vijk \ Vijk is value at k), l < i < w, l < j < h, l < k < b) (2) w = (y ^ ijk \ ^ ijk is value at p (z, < i < w, l < j < h, l < k < b} (3) First, the function of point processing for each band pixel 値 of the aforementioned image will be described below. When the image X is converted into a binary image, the band pixels 像素 are binarized according to Equation 4. Printed clothes by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs ΦύΑ: (χ) ⑷ 1 if xijk > 〇, L 0 otherwise. When the image x is converted into the image with the largest frequency band, according to Equation 5, from row j column For each band of pixels, select the largest one. In addition, since the frequency band maximum chirp image is a single-band image, it is processed as the image with a frequency band number 1 for convenience. Therefore, the third addition of the function BuKx) is 66. This paper size is applicable to the Chinese National Standard (CNS) A4 specification (10 X 297 mm) 540010. Printed by A7 of the Intellectual Property Bureau Staff Consumer Cooperatives of the Ministry of Economic Affairs. 5. The description of the invention (β) is 1. ^ 71 (x) = max {xijk} ⑸ k image x 舄 2 値 image, when the image χ is inverted, it is calculated according to Equation 6. ^ ijk {x) = 1 — Xijk (6) The logarithmic transformation in the position P (1, j, k) of the image χ is performed according to Equation 7. In this formula, e is the offset, which is used in order to make the output of the natural logarithmic function 値 in the valid range. Generally speaking, e == 1. By using this logarithm, it is possible to divide the band pixels into subtraction. In addition, when the image X is a 2n-tone digital image, it has nothing to do with the number of frequency bands. If there is a list of 2n elements in the memory, it is not necessary to calculate the natural logarithmic function every time. Logarithmic table.
Lijk(x) == ln(a^mA: + e) · ⑺ 另外’圖像的位置p(i、」、k)的q附近位置的集合 P#(Q)疋以式8來顯示。但是，8q是4、8、24、48、80、 120、（2r+ I)2 — 1的連續序列，r是自然數。又，當超出圖 像尺寸的位置包含於集合puk(q)中時，即以無特別限定的位 置P(i、j、k)加以取代。又，除此之外時，即根據指定，易 像素値相當於〇、且圖像所不包含的架空位置加以取代。 據此，Bb自動地進行邊緣處理。因此，集合puk(Q)的要素數 經常爲q。 一 67Lijk (x) == ln (a ^ mA: + e) · ⑺ In addition, the set of positions near q of the image position p (i, ", k) P # (Q) 疋 is shown by Equation 8. However, 8q is a continuous sequence of 4, 8, 24, 48, 80, 120, (2r + I) 2 — 1, and r is a natural number. When a position exceeding the image size is included in the set puk (q), it is replaced by a position P (i, j, k) without any particular limitation. In other cases, by designation, the easy pixel。 is equivalent to 0, and the overhead position not included in the image is replaced. Accordingly, Bb automatically performs edge processing. Therefore, the number of elements in the set puk (Q) is often q. One 67
本紙張尺度適用中國國家標準（CNS)A4 ϋ《210 X 297公H (請先閱讀背面之注意事項再填寫本頁) |裝--------訂---------線. 540010 A7 B7 五 、發明說明 {p(i + U^), p{ij + 1,/c), p{i - lj,k), p{ij - l^k)} if 7 47 {p{l,rrhk) \i — r<l<i + r、j — r <m'<j + r, if「（2 厂+ 1)2- ⑻ 接下來，就有關對圖像的各頻帶圖像値之最大8附近 的附近處理的函數及運算子’說明如下。 圖像X的位置P(1、j、k)的振動是根據式9來進行。因 此，從位置P(i、j、k)的Q附近中，藉只選擇一個位置的方 法，即能決定用圖像單位加以振動或用像素單位加以振雲力 。若在圖像X的所有位置，以完全相同的方法從Q附近中 選擇一個的話，圖像X即以圖像單位振動。另一方面’若 使用擬似隨機數等’隨機地從Q附近中選擇一個的話’ ® 像X即以像素單位振動。 k(x) = ximk for only one of p(l, m, k) e PijkiQ) ⑼ - -----------#裝--------訂---------線· (請先閱讀背面之注意事項再填寫本頁) 經濟部智慧財產局員工消費合作社印製 圖像X的位置p(i、:i、k)的平滑化是根據式10來進行 。但是，int(v)係捨去實數V的小數點以下之意。若圖像X 的頻帶像素値是整數値的話，在組裝硬體時，藉變更電路 以使N#=4時對Xlmk的總和執行2次右偏移指令’ Nuk^8 時對xw的總和執行3次右偏移指令，即能省略執行相除 的電路。 > : ^Imk) (ίο) 68 本紙張尺度適@ ^?TCNS)A4規格《210 χ 297公爱） 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 ------ 五、發明說明 雖是拉普拉斯運算碼的計算，但這是如式11所示，僅 係2階差分運算子。在8的附近，捕捉雜訊的微妙變化零 點及零交叉變多，故適合本發明。但，由於Nuk是4或8， 因此若組裝硬體時，變更電路，以使Nuk=4時對xuk執行2 次左偏移指令，Nuk= 8時對xw執行3次左偏移指令，即能 省略實數之乘算的電路。 ▽ijA：X = 工Irak 一 ^ijk^ijk (11) p(l，m，k)6Pijk(q) 根據利用拉普拉斯運算碼所求得的値，就尋找零點的 方法而言，以往係尋找從正變化到負的像素，但本發明是 根據式12，除了從負到正進行零交叉外的像素外，亦尋找 從負到零或從零到正等經由零點、或持續零的像素。本發 明，式12所尋找到的零點，並非在有邊緣的位置，而是在 有雜訊的位置，亦即位於沒有邊緣的位置。又，亦同時根 據式12進行實數値的二値化。 、ί 1 if xijk < 〇 and xlmk > 〇 for m, k) € Pijk{q)^ I 0 otherwise. 就圖像X是任意的二値圖像而言，當彌補在圖像X中 空孔的像素時，係根據式13進行計算。此處，f是顯示應 彌補孔大小的參數，一般說來，f=l即可。又，4附近時 ’由於無法檢測其性質上的對角線，故儘量在8附近較佳 〇 * 69 本紙張尺度適用中國國家標準（CNS)A4規格（210x297公釐） (請先閱讀背面之注意事項再填寫本頁)This paper size applies to China National Standard (CNS) A4 210 "210 X 297 male H (Please read the precautions on the back before filling this page) | Install -------- Order -------- -Line. 540010 A7 B7 V. Description of the invention {p (i + U ^), p (ij + 1, / c), p (i-lj, k), p (ij-l ^ k)} if 7 47 {p {l, rrhk) \ i — r < l < i + r, j — r < m '< j + r, if "(2 factory + 1) 2- ⑻ The functions and operators' in the vicinity of the maximum 8 near each band image 说明 are described below. The vibration of the position P (1, j, k) of the image X is performed according to Equation 9. Therefore, from the position P ( (i, j, k) near Q, by selecting only one position, you can decide to use the image unit to vibrate or the pixel unit to vibrate the cloud. If all positions of the image X, the exact same Method If one is selected from the vicinity of Q, the image X is vibrated in units of images. On the other hand, 'if a pseudo-random number is used, etc., one is randomly selected from the vicinity of Q' ® The image X is vibrated in pixels. K (x) = ximk for only one of p (l, m, k) e PijkiQ) ⑼------------ # 装 -------- --------- Line · (Please read the notes on the back before filling this page) Smoothing the position p (i,: i, k) of the image X printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs The conversion is performed according to Equation 10. However, int (v) is rounded down to the decimal point of the real number V. If the band pixel 値 of the image X is an integer ，, when the hardware is assembled, the circuit is changed so that When N # = 4, the right offset instruction is executed twice for the sum of Xlmk. 'Nuk ^ 8 When the right offset instruction is executed for the sum of xw 3 times, the division circuit can be omitted. ≫: ^ Imk) (ίο ) 68 paper size @ ^? TCNS) A4 specification "210 χ 297 public love" Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 540010 A7 B7 ------ 5. Although the description of the invention is Laplace operation The calculation of the code, but as shown in Equation 11, is only a second-order difference operator. In the vicinity of 8, the subtle changes of the captured noise have more zeros and zero crossings, so it is suitable for the present invention. However, since Nuk is 4 Or 8, if the hardware is assembled, change the circuit so that Nuk = 4 executes 2 left shift instructions to xuk, and Nuk = 8 executes 3 left shift instructions to xw, ie Circuits that can multiply real numbers. ▽ ijA: X = Irak ^ ijk ^ ijk (11) p (l, m, k) 6Pijk (q) According to 値 obtained using the Laplacian operation code, As far as the method of finding the zero point is concerned, in the past, it is looking for pixels that change from positive to negative, but the present invention is based on Equation 12, in addition to pixels that make a zero crossing from negative to positive, it also looks for negative to zero or from zero to Waiting for pixels passing through zero, or continuing to zero. In the present invention, the zero point found in Equation 12 is not a position with edges, but a position with noise, that is, a position without edges. At the same time, binarization of real numbers is performed according to Equation 12. , Ί 1 if xijk < 〇and xlmk > 〇for m, k) € Pijk (q) ^ I 0 otherwise. As far as image X is an arbitrary binary image, when the hole in image X is made up Pixel is calculated according to Equation 13. Here, f is a parameter showing the size of the hole that should be compensated. Generally speaking, f = 1. Also, when it is near 4 ', it is better to be near 8 because it is impossible to detect the diagonal of its nature. * 69 This paper size applies the Chinese National Standard (CNS) A4 specification (210x297 mm) (Please read the back (Please fill in this page again)
540010 A7 B7 五 、發明說明（tJ) =< 1 H ^2p(l，m，k)GPij以q)工Imk + f k Nijk， Xijk otherwise. (13) 就圖像X是任意的二値圖像而言，當削除圖像X中孤 立點或孤立孔時，係根據式14進行計算。又，4附近時， 由於無法檢測其性質上的對角線，故儘量在8附近較佳。 =< 0 if Xijk — 1 3-Ild ^lp(l^m,k)^Pijk(g) — 〇， 1 if 工⑽=〇 and ⑷而爪无=M#， (14) Xijk otherwise. (請先閱讀背面之注意事項再填寫本頁) 裝 就圖像X是任意的二値圖像而言，爲了檢測圖像X中 線寬爲1的像素，係使用4附近像素，根據式15進行計算540010 A7 B7 V. Description of the invention (tJ) = < 1 H ^ 2p (l, m, k) GPij, q), Imk + fk Nijk, Xijk otherwise. (13) As far as the image X is an arbitrary binary map For example, when an isolated point or an isolated hole in the image X is deleted, it is calculated according to Equation 14. Also, in the vicinity of 4, it is not possible to detect a diagonal line in nature, so it is better to be in the vicinity of 8. = < 0 if Xijk — 1 3-Ild ^ lp (l ^ m, k) ^ Pijk (g) — 〇, 1 if 工 ⑽ = 〇and ⑷ and the claw is not = M #, (14) Xijk otherwise. ( Please read the precautions on the back before filling in this page.) Assume that image X is an arbitrary binary image. To detect pixels with a line width of 1 in image X, use 4 nearby pixels according to Equation 15. Calculation
Jijk (^) 工ijk if Xi^ijk Xi^ijk — 0 ΟΓ 一 〇， (15) 經濟部智慧財產局員工消費合作社印製 0 otherwise. 若兩個圖像x、y是任意的二値圖像，圖像y係檢測圖 像X中線寬爲1之像素的話，爲了擴充圖像X中線寬爲1 的像素線寬，係使用4附近像素，根據式16進行計算。 1 if 仏一ljA: + W+ljfc 十 Vij·- 1A: + J/ij+U〉◦) Xijk otherwise. (16) 因此，若使用式15的線寬檢知與式16的線寬擴充的 話，根據式17，即能簡單地記述二値圖像的線寬內插。Jijk (^) 工 ijk if Xi ^ ijk Xi ^ ijk — 0 ΟΓ 10, (15) Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economy 0 otherwise. If the two images x and y are arbitrary two-dimensional images If the image y is a pixel with a line width of 1 in the detection image X, in order to expand the line width of the pixel with a line width of 1 in the image X, a pixel near 4 is used and calculated according to Equation 16. 1 if 仏 一 ljA: + W + ljfc V Vij ·-1A: + J / ij + U〉 ◦) Xijk otherwise. (16) Therefore, if the line width detection of formula 15 and the line width expansion of formula 16 are used According to Equation 17, the line width interpolation of the second image can be simply described.
Cijk(^) = J(x)) (17) 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 540010 A7 B7 經濟部智慧財產局員工消費合作社印製 五、發明說明（ 其次，就有關對圖像之各頻帶像素値之附近處理的函 數及運算子，說明如下。 當有兩個圖像X、y時，該等圖像的最大値圖像是根據 式18來計算。 if Xijk ^ Vijk) yijk otherwise. (18); 當有兩個圖像x、y時，這些屬像的差分是根據式19 來計算。 (19) 此處，若使用式11的算法與式19的差分的話，即能 根據式20，簡單地記述圖像的淸晰化。 = Dijk^ V^X) (20) 當有兩個圖像χ、y，圖像y是單頻帶二値圖像時，即 能根據式21，使用圖像y的頻帶像素値函蓋圖像X各頻帶 像素値。 = Xijkyiji (21) 當有兩個圖像x、y，圖像x與圖像y是二値圖像時 即能根據式22，以圖像X爲基準對圖像y進行整形。 71 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 丨 1---·------·裝 (請先閱讀背面之注意事項再填寫本頁) 訂--------- 540010 Α7 Β7 經濟部智慧財產局員工消費合作社印製 五、發明說明（ ^ijk if + Σρ(Ζ，?τι，Α:)€/^Α：⑷奶> 0, 0 otherwise. (22) 當有兩個圖像x、y，圖像y是二値圖像時，根據式23 ，在圖像X的頻帶像素値附近中，用圖像y所指定圖像X 的頻帶像素値的平均値，內插圖像y所未指定的圖像X的 頻帶像素値。但是，int(v)係捨去實數V的小數點以下之意Cijk (^) = J (x)) (17) This paper size applies to the Chinese National Standard (CNS) A4 (210 X 297 mm) 540010 A7 B7 Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs Next, the functions and operators related to the processing of the pixels around each band of the image are described below. When there are two images X and y, the maximum image of these images is calculated according to Equation 18. If Xijk ^ Vijk) yijk otherwise. (18); When there are two images x, y, the difference between these images is calculated according to Equation 19. (19) Here, if the difference between the algorithm of Equation 11 and Equation 19 is used, the sharpening of the image can be simply described according to Equation 20. = Dijk ^ V ^ X) (20) When there are two images χ, y, and the image y is a single-band two-frame image, the image can be covered using the frequency band pixels of the image y according to Equation 21. Pixels in each frequency band X. = Xijkyiji (21) When there are two images x, y, and the images x and y are two-dimensional images, the image y can be shaped according to Equation 22 with the image X as the reference. 71 This paper size is applicable to China National Standard (CNS) A4 specification (210 X 297 mm) 丨 1 --------------- (please read the precautions on the back before filling this page) Order- ------- 540010 Α7 Β7 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs V. Invention Description (^ ijk if + Σρ (Z,? Τι, Α :) € / ^ Α: ⑷ 奶 > 0, 0 otherwise. (22) When there are two images x, y, and the image y is a two-dimensional image, according to Equation 23, in the vicinity of the frequency band pixels 图像 of the image X, the image X specified by the image y The average 値 of the frequency band pixels, interpolates the frequency band pixels 图像 of the image X not specified by the image y. However, int (v) means that the real number V is rounded down to the decimal point.
Vijki^y) =< int(Vijki ^ y) = < int (
Xijk ^ρα,Ηΐ,ΐ)βρ〇1(9)yiml if yiji = 0 and Ep(/,ma)€p..l(,) yimi > otherwise. (23) 又，本發明中，像素位置與移動量等亦像圖像資料般 單純地加以處理。這種處理叫做位置的圖像化。以下，就 有關圖像化的幾個函數及運算子，加以說明。 首先，將位置p(l、m、〇)的1、m、〇各値作爲圖像資 料，將轉換爲頻帶像素値的運算子作爲#，將轉換的頻帶像 素値作爲#?(1、m、〇)。其次，考慮頻帶像素値從位置p(: 、j、k)移動到位置p(i + 1、j + m、k + 〇)的情形。此時，頻 帶像素値的移動量係以p(l ' m、〇)加以表示。亦即，可將 移動量視爲來自某位置的向量。最後，將從頻帶像素値取 出位置的運算子作爲#^。因此，#-Wpd、m、〇)= P(1、tr 、ο) 〇 此處，根據式24，可在以寬度方向與高度方向所顯牙 72 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） -----------裝---I---"訂--------- (請先閱讀背面之注意事項再填寫本頁) 540010 A7 B7 經濟部智慧財產局員工消費合作社印製 五、發明說明（/)丨） 的平面內，將移動量p(i、j、k)朝向180度相反方向 k)) -7, k) (24) 當有圖像X，圖像X是單頻帶二値圖像時，圖像X之 位置p(i、j、k)的重心位置移動量是根據式25來計算。又 ，當計算本來重心時雖須進行除算，但由於計算有關‘8附 近內的移動量時，除算被抵消，故式25省略了除算。 (卜如 ml， Σ (m ~ j)^mi,〇) p(l,m,l)ePlji{q) p(l}myl)ePiji(q) (25) 從移動量p(i、j、k)，根據式26及27，計算有關至 附近內的移動量，即能使移動量圖像圖像化。又，式27 僅係利用在使用圖像的離散化用式26無法對應時。 Q{p{hj,k)) {Jp(l，0，A:) if 2 > 0, \j\ < |ζ|/2, 胗(1，一1，幻 if2>0,j<0, W/2<|j|<2|z|, 眇(0, 一 1，々）ifj <0,2K|<|儿 眇(一 1，一 1，A:) if 2 <0,；<0, \i\/2<\j\<2\il 眇(一 1，0，岣 if2<0, \j\ < HA (—1，1，々）ifz<0,i>0, N/2< |j| < 2|z|, iip(0，l，A:) ifj>0,2H<|儿鉍(1，1，） if2>〇,j>〇, N/2<|i|<2|z|, |jp(0, 0, k) otherwise. (26) ------------裝--------訂--------- (請先閱讀背面之注意事項再填寫本頁) 73 本紙張尺度適用中國國家標準（CNS)A4規格¢210 X 297公爱） 540010 A7 B7 五、發明說明工) ^(1,0,^) if i > 0, \j\ < |i|/2, 眇(1，0,々）if i > 0, j < 〇, \i\/2 < \j\ < 2|2f5 Up(07 1，A:) if i < 0, j > 〇, \i\/2 < \j\ < 2|z|, ftp(0，l，A〇 if; > 0, 2|i| < \jl jjp(l, 1, k) if z > 0, j > 〇7 |^|/2 < \j\ < 2|2|? ||p(0, 0, k) otherwise. 因此，若使用式25、26及27的話，根據式28及29， 能簡單地記述有關單頻帶二値圖像X之向重心方向的移動 量圖像的頻帶像素値。又，移動量圖像的頻帶數爲1。 = (27) Δ^·ι(χ) = ©(G^i(x)) Δί；ι(χ) = Θ，(^ι(χ)) (28) (29) 另一方面，由於使用式24能求出重心位置的相反位置 ，故根據式30，能簡單地記述有關單頻帶二値圖像X之重 心與向相反方向的移動量圖像的頻帶像素値。又，移動量 圖像的頻帶數爲1。 (30) --U----r-------------^--------- (請先閱讀背面之注意事項再填寫本頁) 經濟部智慧財產局員工消費合作社印製 當有兩個圖像X、y，圖像y是移動量圖像時，根據式 31，將圖像X的頻帶像素値移動至圖像y所指示的移動位 置後，能將移動至相同頻帶像素之頻帶像素値的總和作爲 爲濃淡圖像。 y) = E for ρ(Ζ，τη，1) 6 and jj ^mi = P(i — Z，j — τπ，0)· (3工） 此處，使用式4、28、29及31，根據式32或33，將 74 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 540010 A7 B7 經濟部智慧財產局員工消費合作社印製 77 發明說明（Ίί) 單頻帶濃淡圖像X移動到附近的重心方向後，即能簡單地 記述移動至相同頻帶像素之頻帶像素値的總和。 = Γ^·ι(χ, Δ(Φ(χ)))Xijk ^ ρα, Ηΐ, ΐ) βρ〇1 (9) yiml if yiji = 0 and Ep (/, ma) € p..l (,) yimi > otherwise. (23) In the present invention, the pixel position And the amount of movement and so on are processed simply as image data. This process is called location visualization. In the following, several functions and operators related to visualization are described. First, each of 1, m, and 0 at positions p (l, m, 0) is used as image data, an operator converted into a band pixel 値 is #, and a converted band pixel 値 is #? (1, m , 〇). Next, consider a case where a band pixel 値 moves from a position p (:, j, k) to a position p (i + 1, j + m, k + 0). At this time, the amount of movement of the band pixel 値 is expressed by p (l'm, 0). That is, the amount of movement can be considered as a vector from a certain position. Finally, the operator from the position of the band pixel 値 is taken as # ^. Therefore, # -Wpd, m, 〇) = P (1, tr, ο) 〇 Here, according to Equation 24, the teeth can be displayed in the width direction and the height direction. 72 The paper size applies the Chinese National Standard (CNS) A4. Specifications (210 X 297 mm) ----------- install --- I --- " order --------- (Please read the precautions on the back before filling in this Page) 540010 A7 B7 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 5. In the plane of the invention description (/) 丨), move the movement amount p (i, j, k) in the opposite direction of 180 degrees k)) -7, k) (24) When there is an image X, and the image X is a single-band two-frame image, the shift amount of the position of the center of gravity of the position p (i, j, k) of the image X is calculated according to Equation 25. In addition, although the division must be performed when calculating the original center of gravity, since the division is cancelled when calculating the amount of movement in the vicinity of '8, the division is omitted in Equation 25. (Such as ml, Σ (m ~ j) ^ mi, 〇) p (l, m, l) ePlji (q) p (l) myl) ePiji (q) (25) From the moving amount p (i, j, k) According to Equations 26 and 27, calculate the amount of movement to the vicinity, that is, the image of the amount of movement can be imaged. Equation 27 is used only when Equation 26 for discretization using an image cannot be used. Q (p {hj, k)) {Jp (l, 0, A :) if 2 > 0, \ j \ < | ζ | / 2, 胗 (1,1, magic if2 > 0, j < 0, W / 2 < | j | < 2 | z |, 眇 (0, one 1, 々) ifj < 0,2K | < | er 眇 (one 1, one 1, A :) if 2 <0,; < 0, \ i \ / 2 < \ j \ < 2 \ il 眇 (-1,0, 岣 if2 < 0, \ j \ < HA (-1,1, 々) ifz < 0, i > 0, N / 2 < | j | < 2 | z |, iip (0, l, A :) ifj > 0,2H < | child bismuth (1,1,) if2 > 〇, j > 〇, N / 2 < | i | < 2 | z |, | jp (0, 0, k) otherwise. (26) ------------ install ------- -Order --------- (Please read the notes on the back before filling out this page) 73 This paper size is applicable to Chinese National Standard (CNS) A4 specifications ¢ 210 X 297 public love) 540010 A7 B7 V. Invention Explanation) ^ (1,0, ^) if i > 0, \ j \ < | i | / 2, 眇 (1,0,0) if i > 0, j < 〇, \ i \ / 2 < \ j \ < 2 | 2f5 Up (07 1, A :) if i < 0, j > 〇, \ i \ / 2 < \ j \ < 2 | z |, ftp ( 0, l, A〇if; > 0, 2 | i | < \ jl jjp (l, 1, k) if z > 0, j > 〇7 | ^ | / 2 < \ j \ & lt 2 | 2 |? || p (0, 0, k) otherwise. Therefore, if Equations 25, 26, and 27 are used, according to Equation 28 29. It is possible to simply describe the band pixels 値 of the single-band secondary image X in the direction of the center of gravity of the moving amount image. In addition, the number of frequency bands of the moving amount image is 1. = (27) Δ ^ · ι (χ ) = © (G ^ i (x)) Δί; ι (χ) = Θ, (^ ι (χ)) (28) (29) On the other hand, since the opposite position of the center of gravity can be obtained by using Equation 24, Therefore, according to Equation 30, the center of gravity of the single-band secondary frame image X and the band pixels 图像 of the moving image in the opposite direction can be simply described. The number of frequency bands of the moving image is 1. (30)- U ---- r ------------- ^ --------- (Please read the notes on the back before filling out this page) Employees ’Cooperatives, Intellectual Property Bureau, Ministry of Economic Affairs When there are two images X, y, and the image y is a moving image, according to Equation 31, after moving the band pixel 频带 of the image X to the moving position indicated by the image y, the image can be moved to The sum of band pixels 値 of pixels in the same band is regarded as a gradation image. y) = E for ρ (Z, τη, 1) 6 and jj ^ mi = P (i — Z, j — τπ, 0) · (3 operations) Here, using formulas 4, 28, 29, and 31, according to Formula 32 or 33, 74 paper sizes are applicable to Chinese National Standard (CNS) A4 (210 X 297 mm) 540010 A7 B7 Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 77 Description of the invention After X moves to the direction of the center of gravity nearby, the total sum of the band pixels 値 that move to the pixels in the same band can be simply described. = Γ ^ · ι (χ, Δ (Φ (χ)))
Hjl ㈨二 Γν(χ，Δ，(Φ㈤)） (32) (33) 當有兩個圖像X、y，圖像X爲二値圖像，圖像y爲移 動量圖像時，由於能求出圖像X之各頻帶像素値之移動端 的位置，故能找出移動端重複的頻帶像素値。因此，圖像 X之各頻帶像素値的移動端不至重複，且，能顯示移動之 各頻帶像素値存在之能移動圖像的頻帶像素値，是根據式 34生成。又，能移動圖像的頻帶數爲1。 1 if Xiji = 0 and = p{i -IJ - m, 0) for only one of p(l, m, 1) G Pijiiq), 0 otherwise. (34) 當有三個圖像x、y、w，圖像x爲能移動圖像，圖像 w爲移動量圖像時，能根據式35，移動圖像x的頻帶像素 値。 ximk if Viji = 1 and = p(i - l, j - m, 0) for 3p(l1m1l) G Pl3i{q), TijA:(x，y，w) = < 〇 if Vlrnl — 1 = p(/ — i, m — j, 0) for 3p(/,m, 1) € Pl3i{q), xZJk otherwise. (35) —f — -------. — I----^--------1 (請先閱讀背面之注意事項再填寫本頁) 75 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 540010 五、 經濟部智慧財產局員工消費合作杜印製 A7 B7 --- ^------ 發明說明 此處’右使用式30、34及35的話，根據式36，在與 二値圖像y所計算的重心位置相反方向，能移動圖像χ的 頻帶像素，簡單地記述所得到的圖像的續帶像素値。 uijk(^ y) = Tijkix, H(y, R(y)), R(y)) (36) 使用式1到式36 ’能記述組裝圖像記憶機構12(參照 圖1及圖6)、圖像振動機構13(參照第3_|、第4圖及第5 圖）、邊緣資訊生成機構14(參照第1圖及第6圖）、邊緣畜 s只形成機構15(參照弟1〜5圖、弟7〜12圖）、物|§/背學分 离隹機構16(參照桌2、4、5、7〜12圖）、位置/大小檢測j幾構 17(參照弟1圖及弟6圖）、領域規格化機擒27(參照第7圖) 、及規格化圖像保持機構28(參照第7 _)的資料處理裝置 110的所有陣列運算單元40的算法。以下，使用資料 裝置110中的任意陣列運算單兀40的算法，就對應圖像振 動機構13、邊緣資訊生成機構14、邊緣資訊形成機構15 、位置/大小檢測機構17及領域規格化機構27，從申請專 利範圍第6項到第10項之視覺裝置，加以說明。 資料處理裝置110所實現的圖像記憶機構12(參照第工 圖及第6圖)爲了記憶數位圖像in，網格狀排列的陣列運 算單元40係同步並列地動作。假設格子上1行」列所配置 的陣列運算單元40爲ΑΟΙΙϋ，則AOIL的算法如第14圖所 示。 在步驟1201中，將A〇Uu配置在格子上的1行j列。 無論在邏輯上或物理上，此係爲了決定A0UU的附近所需 者。 76 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） ---------------^裝--------訂---------線« C請先閱讀背面之注意事項再填寫本頁) 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明（7?) 在步驟1202，設定AOUu的附近與變數的初始値。 在步驟1203中，判斷依序輸入的數位圖像111是否已 經沒有。若無數位圖像H1的話(步驟1203 : YES)，即結束 算法。若有數位圖像111的話(步驟1203 : NO)，即移到步 驟1204。但，當僅對特定的圖像尺寸進行組裝陣列運算單 元40時，即使是無限回路亦可。 在步驟1204中，到數位圖像111準備完成止，進行輸 入等待。 在步驟1205中，輸入數位圖像1Π的i行j列像素的 頻帶數部分。因此，AOUu至少需要有記憶頻帶數部分的圖 像資料的記憶體42。 在步驟1206中，爲了在輸入等待期間能夠進行輸出， 係記億數位圖像111的1行j列的像素。 在步驟1207中，輸出數位圖像111的頻帶像素値。之 後，回到步驟1203。 據此，使用由陣列運算單元40所構成的資料處理裝置 110，對應圖像記憶機構12的視覺裝置即能記憶數位圖像 111。 利用資料處理裝置110所實現的申請專利範圍第6項 之圖像振動機構13(第3圖、第4圖及第5圖），爲了使數 位圖像111振動，網格狀排列的陣列運算單元40係同步並 列動作。設網格狀排列的陣列運算單元40爲AOUu，貝Ο A〇Uu的算法如第15圖所示。 在步驟1301中，將ΑΟΙΙϋ配置在格子上的i行j列。 77 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） -------------裝---------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 540010 A7 B7 五、發明說明（Ί “） 無論在邏輯上或物理上，此係爲了決定A〇ui」的附近所需 者。 (請先閲讀背面之注意事項再填寫本頁) 在步驟1302中，設定AOUu的附近與變數的初始値。 在步驟1303中，判斷依序輸入的數位圖像Π1是否已 經沒有。若無數位圖像111的話(步驟1303 ·· YES)，即結束 算法。若有數位圖像111的話(步驟1303 : NO)，即移到步 驟1304。但，當僅對特定的圖像尺寸進行陣列運算單元40 時，即使是無限回路亦可。 在步驟1304中，輸入數位圖像111的i行j列的像素 的頻帶數部分。因此，AOUu至少需要有記億頻帶數部分的 圖像資料的記憶體42。 在步驟1305中，根據函數Ξ#(χ)，將圖像111的1行J 列的圖像移動到附近圖像的一個像素。 在步驟1306中，輸出數位圖像111的頻帶像素値。之 後，回到步驟1303。 據此，使用由陣列運算單元40所構成的資料處理裝® 110，對應圖像記憶機構13的申請專利範圍第6項之視覺 裝置，即能使數位圖像111振動。 經濟部智慧財產局員工消費合作社印製 利用資料處理裝置110所實現的申請專利範圍第7項 之邊緣資訊生成機構14(參照第1圖及第6圖），爲了從數 位圖像111生成粗邊緣資訊圖像113,網格狀排列的陣列 運算單元40係同步並列動作。設格子上i行」列所排列的 陣列運算單元40爲AOUu，則對應邊緣資訊生成機構14的 A〇Uu算法如第16圖所示。 78 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 540010 經濟部智慧財產局員工消費合作社印製 A7 B7___ 五、發明說明（Ή) 在步驟1401中，將ΑΟΙΙϋ配置在格子上的i行j列。 無論在邏輯上或物理上，此係爲了決定A0UU的附近所需 者。 在步驟1402中，設定AOUu的附近與變數的初始値。 在附近的設定中，也可個別地決定前述各函數所使用的附 近大小q爲4或8，也可將全部統一爲4或8。本發明的邊 緣資訊生成機構14，爲了提高所生成的粗邊緣資訊112的 正確度，藉控制計算時間與數位圖像111的頻帶數等，邊 緣資訊生成機構14可視需要改變適當附近大小來加以處置 〇 在步驟1403中，判斷數位圖像111是否已結束。若無 數位圖像111的話(步驟1403 : YES)，即結束算法。若有數 位圖像111的話(步驟1403 : NO)，亦結束算法。但，當對 特定的頻帶數與圖像尺寸，組裝陣列運算單元40時，即使 是無限回路亦可。 在步驟1404中，輸入數位圖像111的i行j列像素的 頻帶數部分。這是因爲AOlLj係一次性的處理數位圖像in 的1行Ί列的像素的關係。因此，A〇仏至少需要有記憶頻 帶數部分的圖像資料的記憶體42。 在步驟1405中，A〇Uu利用與附近的陣列運算單元40 之通信，對所輸入之數位圖像111的各頻帶像素値，根據 Sm(x)進行平滑化。已平滑化的頻帶像素値被作爲平滑化圖 像的頻帶像素値加以處理。此處，函數Sm(x)也可視需要反 覆數次。一般的彩色圖像時，2次即足夠。 79 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐) "一 ------------·裝--------訂---------線ΦΓ (請先閱讀背面之注意事項再填寫本頁) 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（y) 在步驟1406中，對平滑化的各頻帶像素値，根據函數 Luk(x)進行對數轉換。已進行對數轉換的頻帶像素値被作胃 對數轉換圖像的頻帶像素値加以處理。 在步驟1407中，AOUu利用與附近的陣列運算單元40 之通信，對對數轉換的各頻帶像素値，根據Ε#00進行'凊 晰化。已淸晰化的頻帶像素値被作爲淸晰化圖像的頻帶像 素値加以處理。 在步驟1408中，對淸晰化圖像的各頻帶像素値’根據 函數DWx，y)，減去輸入1前之淸晰化圖像的各頻帶像素 値。計算差分的頻帶像素値被作爲時間差分圖像的頻帶像 素値加以處理。 在步驟1409中，用淸晰化圖像所對應的各頻帶像素値 ，取代輸入1前之淸晰化圖像的各頻帶像素値。 在步驟1410中，AOUu是利用與附近的陣列運算單元 40之通信，對時間差分圖像的頻帶像素値，根據運算子 ▽2以X進行拉普拉斯運算碼的計算。計算拉普拉斯運算碼的 頻帶像素値，被作爲時間差分拉普拉斯運算碼圖像的頻帶 像素値加以處理。 在步驟1411中，AOlLj是利用與附近的陣列運算單元 40之通信，對時間差分拉普拉斯運算碼圖像的各頻帶像素 値，根據函數Ζ4χ)抽出零點。抽出零點的頻帶像素値被 作爲時間差分零點圖像的頻帶像素値加以處理。 在步驟1412中，對時間差分拉普拉斯運算碼圖像的各 頻帶像素値，根據函數Bm(x)在各頻帶像素中檢測最大値 80 本紙張尺度適用中國國家標準（CNS)A4規格(210 X 297公釐） " ------------裝--------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（1) 。經檢測的最大値頻帶像素値，被作爲最大値時間差分零 點圖像的頻帶像素値加以處理。又，爲方便起見，頻帶數 爲1。 在步驟1413中，AOUu是利用與附近的陣列運算單元 40之通信，對淸晰化圖像的各頻帶像素値，根據根據運算 子進行拉普拉斯運算碼的計算。計算拉普拉斯運算碼 的頻帶像素値，被作爲時間差分拉普拉斯運算碼圖像的頻 帶像素値加以處理。 在步驟1414中，AOUu是利用與附近的陣列運算單元 40之通信，對拉普拉斯運算碼圖像的各頻帶像素値，根據 函數Ζμ(χ)抽出零點。抽出零點的頻帶像素値被作爲時間 差分圖像的頻帶像素値加以處理。 在步驟1415中，對拉普拉斯運算碼圖像的各頻帶像素 値，根據函數BuKx)在各頻帶像素中檢測最大値。經檢測 的最大値頻帶像素値，被作爲最大値時間差分零點圖像的 頻帶像素値。又，爲方便起見，頻帶數爲1。 在步驟1416中，對拉普拉斯運算碼圖像的各頻帶像素 値與時間差分拉普拉斯運算碼圖像的各頻帶像素値，根據 函數M#(x，y)在位於各圖像相同位置的頻帶像素値中，檢 測最大値。經檢測的最大頻帶像素値，被作爲混合零點圖 像的頻帶像素植。又，爲方便起見，頻帶數爲1。 在步驟1417中，AOUu是利用與附近的陣列運算單元 40之通信，對混合零點圖像的頻帶像素値，根據函數 F#(x)除去孔。經除去孔的頻帶像素値，被作除去混合零點 81 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） ------------·裝--------訂---------線· (請先閱讀背面之注意事項再填寫本頁) 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（p) 圖像的頻帶像素値。又，爲方便起見，頻帶數爲1。此處 ，函數Fm(x)可視需要反覆數次。一般的彩色圖像時，1次 即足夠。 在步驟1418中，人011"是利用與附近的陣列運算單元 40之通信，對除去孔混合零點圖像的頻帶像素値，根據函 數ΑυφΟ除去孤立點及孤立孔。經除去孤立點及孤立孔的 頻帶像素値，被作爲除去雜訊混合零點圖像的頻帶像素値 加以處理。又，爲方便起見，頻帶數爲1。 在步驟1419中，對除去混合零點圖像的頻帶像素値， 根據函數Ιμ(χ)反轉0與1。經反轉的頻帶像素値，被作爲 粗邊緣資訊圖像113的頻帶像素値加以處理。 在步驟1402中，輸出粗邊緣資訊圖像113的頻帶像素 値。之後，回到步驟1403。 因此，使用由陣列運算單元40所構成的資料處理裝置 110，對應邊緣資訊生成機構14的申請專利範圍第7項之 視覺裝置，能從數位圖像111生成粗邊緣資訊圖像113。 如第17圖所示，根據資料處理裝置110所實現的邊緣 資訊形成機構15(參照第1〜5圖，7〜12圖），爲了根據粗 邊緣資訊112所構成的申請專利範圍第8項之粗邊緣資訊 圖像113及數位圖像111，生成由形成邊緣資訊114所構成 的形成邊緣資訊圖像115,網格狀排列的陣列運算單元40 是同步並列動作。設格子上1行」列所配置的陣列運算單 元40爲AOUu，A〇Uu的算法則如第18圖所示。 在步驟1501中，將A〇Uu配置在i行j列。無論在邏 82 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） ---f---〇------裝---------訂 ---------線 (請先閱讀背面之注意事項再填寫本頁) 540010 A7 B7 五、發明說明（yl) 輯上或物理上，此係爲了決定AOUU的附近所需者。 (請先閱讀背面之注意事項再填寫本頁) 在步驟1502中，設定AOUu的附近與變數的初始値。 在附近設定時，可個別決定前述各函數所使用的附近大小 q爲4或8 .，也可將全部統一爲4或8。本發明的邊緣資訊 資訊形成機構15，爲了提局所形成的形成邊緣資訊114的 正確性，將附近大小q全部設定爲8較佳。但，爲了形成 粗邊緣資訊112，利用計算時間的限制與所輸入的數位圖 像111的頻帶數等，邊緣資訊形成機構15，可視需要適當 地變更附近尺寸，以進行處置。 在步驟1503中，判斷依序輸入的數位圖像111或粗邊 緣資訊圖像113是否已經沒有。若數位圖像111或粗邊緣 資訊圖像113兩者皆無的話(步驟1503 : YES)，即結束算法 。若數位圖像111或粗邊緣資訊圖像113兩者皆有的話(步 驟1503 : NO)，即移到步驟1504。但，當對特定的頻帶數 與圖像尺寸組裝陣列運算單元40時，作爲無限回路亦可。 經濟部智慧財產局員工消費合作社印製 在步驟1504中，輸入數位圖像111及粗邊緣資訊圖像 113的i行」列像素的頻帶數部分。這是因爲AOUu爲了一 次性的處理數位圖像111及粗邊緣資訊圖像113的i行j列 的像素的關係。因此，AOIL至少需要記憶頻帶數部分的圖 像資料的記憶體42。 在步驟1505中，分離數位圖像111的1行」列的像素 與粗邊緣資訊圖像113的i行j列的像素。這是因爲A〇Uu 將數位圖像111的1行」列的像素與粗邊緣資訊圖像113的 1行j列的像素，作爲分別獨立的圖像像素加以處理的關係 83 $紙張尺度適用中國國家標準（CNS)A4規格(210 X 297公爱) 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明（^JL) 。若數位圖像111的i行:i列的像素與粗邊緣資訊圖像113 的i行j列的像素，在一開始時即以分離狀態加以輸入的 話，即無需進行任何處理。 在步驟1506中，AOUu利用與附近的陣列運算單元40 之通信，對所輸入的數位圖像111的各頻帶像素値，根據 函數Suk(x)進行平滑化。經平滑化的頻帶像素値，被作爲平 滑化圖像的頻帶像素値加以處理。此處，函數SUk(x)也可視 需要反覆數次。一般的彩色圖像時，2次即足夠。 在步驟1507中，對平滑化圖像的各頻帶像素，根據 Luk(x)進行對數轉換。經對數轉換的頻帶像素値，被作爲對 數轉換圖像的頻帶像素値加以處理。 在步驟1508中，AOlLj是利用與附近的陣列運算單元 40之通信，對像素値根據函數Εϋφ〇進行淸晰化。經淸晰 化的頻帶像素値，被作爲淸晰化圖像的頻帶像素値加以處 理。 · 在步驟1509中，ΑΟΙΙϋ是利用與附近的陣列運算單元 40之通信，對淸晰化圖像的各頻帶像素値，根據操作v2ljkX 進行拉普拉斯運算碼計算。經計算拉普拉斯運算碼的頻帶 像素値，被作爲拉普拉斯運算碼圖像的頻帶像素値加以處 理。 在步驟1510中，AOUu是利用與附近的陣列運算單元 40之通信，對拉普拉斯運算碼圖像的各頻帶像素値，根據 函數Z^(x)抽出零點。經抽出零點的頻帶像素値，被作爲 零點圖像的頻帶像素値加以處理。 84 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） ---.---f I----裝--------訂---------線1^1" · (請先閱讀背面之注意事項再填寫本頁) 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明（浐y 在步驟1511中，對零點圖像的頻帶像素値，根據函數 B^U)在各頻帶像素値中檢測最大値。所檢測的最大頻帶像 素値，被作爲最大値零點圖像的頻帶像素値加以處理。又 ，爲方便起見，頻帶數爲1。 在步驟1512中，對最大値零點圖像的頻帶像素値，根 據函數1^00反轉0與1。經反轉的頻帶像素値，被作爲基 礎邊緣資訊圖像的頻帶像素値加以處理。 在步驟1513中，所輸入的粗邊緣資訊圖像113的頻帶 像素値，最初被作爲整形粗邊緣資訊圖像的頻帶像素値加 以處理，AOUu利用與附近的陣列運算單元40之通信’使 用基礎邊緣資訊圖像的頻帶像素値，對整形粗邊緣資訊® 像的頻帶像素値，根據函數Qw(x，y)進行整形。經整形的 頻帶像素値，再被作爲整形粗邊緣資訊圖像的頻帶像素値 加以處理。此處，函數Q4x，y)係反覆到本來整形粗邊緣 資訊圖像的頻帶像素値不再變化爲止。但，根據計算時^ 的限制、所輸入的粗邊緣資訊圖像113的品質、所形成的 形成邊緣資訊圖像115所求的品質等，整形處理藉適當的 反覆次數來結束算法較佳。 在步驟1514中，AOUu是利用與附近的陣列蓮算單元 40之通信，對整形粗邊緣資訊圖像的頻帶像素値，根據函 數Cuk(x)進行線寬內插。經內插的頻帶像素値，被作爲形 成邊緣資訊圖像115的頻帶像素値加以處理。 在步驟1515中，輸出形成邊緣資訊圖像Π5的頻帶像 素値。之後，回到步驟1503。 85 --i---U------裝--------訂---------線 ^ΙΓ (請先閱讀背面之注意事項再填寫本頁) 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 540010 A7 ____ B7 五、發明說明（#) 因此’使用由陣列運算單元40所構成的資料處理裝置 110，對應邊緣資訊形成機構15的申請專利範圍第8項之 視覺裝置，能生成由粗邊緣資訊圖像113所構成的形成邊 緣資訊圖像115。 此處，所謂從粗邊緣資訊圖像113到形成邊緣資訊圖 像115的形成，可視爲從拍攝某情形之低解像度的數位圖 像111所生成的邊緣資訊，推算拍攝相同情形之高解像度 之數位圖像111所生成的邊緣資訊。此處，對自然數η， 如第19圖所示，根據降低數位圖像U1解像度ι/η之低解 像度數位圖像116，使用邊緣生成機構η生成低解像度粗 邊緣資訊圖像117時，將低解像度粗邊緣資訊圖像U7放 大η倍，即能生成粗邊緣資訊圖像ία。但，爲了將數位 圖像111的解像度設定在1/η，於水平及垂直方向單純地自 數位圖像111的連續η個中抽出1個即可。此外，爲了將 低解像度粗邊緣資訊圖像117放大η倍，於水平及垂直方 向單純地將低解像度粗邊緣資訊圖像117的連續像素間頻 帶像素値爲0的像素彌補η- 1個即可。此時，若η不太大 的話，實現邊緣資訊形成機構15的資料處理裝置11〇，形 成放大低解像度粗邊緣資訊圖像117的粗邊緣資訊圖像 113的形成邊緣資訊圖像115，與實現邊緣資訊形成機構15 的資料處理裝置110，形成由數位圖像111所生成的粗邊 緣資訊圖像113的形成邊緣資訊圖像115，大致相同。其 理由是因爲邊緣資訊形成機構15，在使用數位圖像111於 內部生成的邊緣資訊中使用哪一個邊緣資訊來做爲參考， 86 * 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） (請先閱讀背面之注意事項再填寫本頁) 訂---------線秦 經濟部智慧財產局員工消費合作社印製 經濟部智慧財產局員工消費合作社印製 540010 A7 ________ B7 五、發明說明（&) 邊緣資訊形成機構15僅使用粗邊緣資訊圖像113的關係。 因此，當將放大低解像度粗邊緣資訊圖像117的粗邊緣資 訊圖像113，輸入邊緣資訊形成機構15時，從低解像度數 位圖像116實現生成粗邊緣資訊圖像in的邊緣資訊生成 機構13的資料處理裝置11〇，能減低硬體量。 若進一步發展該方法，如第20圖所示，在降低數位圖 像111之解像度的低解像度數位圖像116所生成的低解像 度粗邊緣資訊圖像117中，能生成切出粗邊緣資訊112周 邊的低解像度切出粗邊緣資訊圖像118。若將放大該低解 像度切出粗邊緣資訊圖像118的切出粗邊緣資訊圖像119 ，與在數位圖像111中切出相同領域的切出數位圖像120 ，輸入實現邊緣資訊形成機構14的資料處理裝置110的話 ，即能生成切出形成邊緣資訊圖像121。此時，實現邊緣 資訊形成機構14的資料處理裝置11〇，能減低硬體量。 如第21圖所示，利用資料處理裝置110來實現的申請 專利範圍第9項之位置/大小檢測機構17(參照第1圖及第6 圖），爲了從將粗邊緣資訊112作爲像素的粗邊緣資訊圖像 113，生成將重複資訊131作爲像素的重複資訊圖像132， 網格狀排列的陣列運算單元40係同步並列地動作。設格子 上i行j列所配置的陣列運算單元40爲A〇Uu，則AOU,』的 算法如第22圖所示。 在步驟1701中，將AOU,」配置在格子上1行j歹fj。此 係理論性或物理性，此係爲了決定A0UU的附近所需者。 在步驟1702中，設定AOUu的附近與變數的初始値。 87 本紙張尺度適用中國國家標準（CNS)A4規格（210x 297公釐） --i---------裝--------訂--------線 (請先閱讀背面之注意事項再填寫本頁) 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（^) 在附近的設定時，可個別設定前述各函數所使用的附近大 小q，亦可將全部統一。本發明的資料處理裝置11〇 ’爲了 提高所生成的重複資訊圖像132的正確性，最好晏能將附 近大小q全部設定較大的値。但，若因爲計算物體的粗邊 緣資訊112的重心所利用計算時間的限制、所輸入的粗邊 緣資訊圖像113的尺寸等’位置/大小檢測機構Π可視需 要適當地變更附近尺寸’以進行處置。 在步驟1703中，判定依序輸入的粗邊緣資訊圖像113 是否已經沒有。若無粗邊緣資訊圖像1Π的話(步驟Π03 : YES)，即結束算法。若有粗邊緣資訊圖像113的話（步驟 1703 : NO)，即移到步驟1704。但，當僅對特定的圖像尺 寸進行組裝陣列運算單元40時，作爲無限回路亦可。 在步驟1704中，輸入粗邊緣資訊圖像113的1行：i列 的像素1頻帶部分。因此，AOlLj至少需要記憶1頻帶部分 的圖像資料的記憶體42。 在步驟1705中，將粗邊緣資訊圖像113的粗邊緣資訊 112轉換爲重複資訊圖像132的重複資訊131。重複資訊 131爲相當於1或0的頻帶像素値。 在步驟1706中，A0U。是利用與附近的陣列運算單元 40之通信，對重複資訊圖像132的各頻帶像素値，根據函 數ΔμΟΟ計算移動量。已將移動量圖像化的頻帶像素値，被 作爲移動量圖像的頻帶像素値加以處理。 在步驟1707中，AOUu是利用與附近的陣列運算單元 40之通信，根據函數Λ,Κχ)使重複資訊圖像132的各頻帶 88 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） ：U------------—^--------- (請先閱讀背面之注意事項再填寫本頁) 540010 A7 B7 五、發明說明（抑） 據此，使用由陣列運算單元40所構成的資料處理裝置 110，對應位置/大小檢測機構17的申請專利範圍第9項之 視覺裝置，能從粗邊緣資訊圖像113生成重複資訊圖像 132。當然，申請專利範圍第9項之視覺裝置，亦能不使用 粗邊緣資訊圖像113，從形成邊緣資訊圖像115生成重複 資訊圖像132。因此，使用由陣列運算單元40所構成的資 料處理裝置110，對應位置/大小檢測機構17的視覺裝置， 能從形成邊緣資訊圖像115生成重複資訊圖像132。 在第22圖之算法中，若使用顯示物體領域141的物體 領域圖像142取代由粗邊緣資訊112所構成的粗邊緣資訊 圖像113的話，如第23圖所示，利用資料處理裝置110所 實現的位置/大小檢測機構17(參照圖2及圖4)，也能從顯 示物體領域141的物體領域圖像142,生成顯示重複資訊 131的重複資訊圖像132。但，當使用物體領域圖像142時 ，重複資訊131的重複資訊圖像132,係以其位置爲中心 的物體領域141的像素總數加以顯示，故其結果，意味著 以其位置爲中心的物體的面積。因此，當從重複資訊圖像 132求物體的大小時，要注意使用重複資訊ι31的平方根 ° 如第24圖所示，利用資料處理裝置no所實現的申請 專利範圍第10項之領域規格化機構27(參照第7圖），爲了 生成包含物體領域141的物體領域圖像142及從數位圖像 111包含規格領域144的規格化圖像145，網格狀排列的陣 列運算單元40是同步並列進行動作。設配置在格子上1行 90 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） (請先閱讀背面之注意事項再填寫本頁) ▼裝·-------訂---------線秦 經濟部智慧財產局員工消費合作社印製 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明（θ) j列的陣列運算單元40爲ΑΟΙΙϋ，則AOUu的算法如第25 圖所示。 在步驟2701中，將AOUu配置在格子上i行」列。此 係理論性或物理性，此係爲了決定AOUij的附近所需者。 在步驟2702中，設定AOU,的附近與變數的初始値。 在附近的設定時，可個別設定前述各函數所使用的附近大 小q，亦可將全部統一。本發明的領域規格化機構27，爲 了提高所生成的規格化圖像145的正確性，最好是能將附 近大小Q全部設定爲較大的値。但，爲了將分離物體領域 143規格化所使用之計算時間的限制、所輸入的數位圖像 111的大小等因素，領域規格化機構27可視需要適當變更 附近尺寸，以進行處置。 在步驟2703中，判斷依序輸入的物體領域圖像142或 數位圖像111是否已經沒有。若無物體領域圖像142或數 位圖像111的話(步驟2703 : YES)，即結束算法。若有物體 領域圖像142或數位圖像111的話(步驟2703 : NO)，即移 到2704。但，當僅對特定的頻帶數及圖像尺寸，組裝陣列 運算單元40時，設定爲無限回路亦可。 在步驟2704中，輸入1頻帶部分物體領域圖像142的 i行列的像素，與頻帶數部分數位圖像111的1行j列的 像素。這是因爲A0%係一次性的處理物體領域圖像142 的i行j列的像素與數位圖像111的1行j列的像素的關係 。因此，AOUu至少需要記憶總頻帶數部分的圖像資料的記 憶體42。 91 --·---:------裝--------訂--------- (請先閱讀背面之注意事項再填寫本頁) 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 540010Hjl ㈨ 二 Γν (χ, Δ, (Φ㈤)) (32) (33) When there are two images X, y, image X is a two-frame image, and image y is a moving image, because The position of the mobile end of each band pixel 値 of the image X is obtained, so that it is possible to find the band pixel 値 where the mobile end overlaps. Therefore, the mobile end of each band pixel 图像 of the image X does not repeat, and the band pixel 値 of the moving image that can display the existence of each band pixel 移动 of the movement can be generated according to Equation 34. The number of frequency bands in which the image can be moved is one. 1 if Xiji = 0 and = p (i -IJ-m, 0) for only one of p (l, m, 1) G Pijiiq), 0 otherwise. (34) When there are three images x, y, and w, When the image x is a moving image, and the image w is a moving amount image, the band pixels 値 of the moving image x can be moved according to Equation 35. ximk if Viji = 1 and = p (i-l, j-m, 0) for 3p (l1m1l) G Pl3i (q), TijA: (x, y, w) = < 〇if Vlrnl — 1 = p ( / — I, m — j, 0) for 3p (/, m, 1) € Pl3i (q), xZJk otherwise. (35) —f — -------. — I ---- ^- ------- 1 (Please read the notes on the back before filling out this page) 75 This paper size applies to China National Standard (CNS) A4 (210 X 297 mm) 540010 5. Employees of the Intellectual Property Bureau of the Ministry of Economic Affairs Consumption cooperation Du printed A7 B7 --- ^ ------ Description of the invention Here, if formulas 30, 34, and 35 are used on the right, according to formula 36, the position of the center of gravity calculated in the second image y is opposite. It is possible to move the band pixels of the image χ, and simply describe the continuation pixels 値 of the obtained image. uijk (^ y) = Tijkix, H (y, R (y)), R (y)) (36) Equations 1 to 36 can be used to describe the assembled image memory mechanism 12 (see Figures 1 and 6), Image vibration mechanism 13 (refer to FIGS. 3_ |, 4 and 5), edge information generation mechanism 14 (refer to FIGS. 1 and 6), edge animal formation mechanism 15 (refer to FIGS. 1 to 5) , Brother 7 ~ 12), object | § / back to study separation mechanism 16 (refer to table 2, 4, 5, 7-12), position / size detection j structure 17 (refer to brother 1 and brother 6 ), The domain normalization machine 27 (refer to FIG. 7), and the algorithms of all the array operation units 40 of the data processing device 110 of the normalized image holding mechanism 28 (refer to FIG. 7 _). In the following, the algorithm using the arbitrary array operation unit 40 in the data device 110 corresponds to the image vibration mechanism 13, the edge information generation mechanism 14, the edge information formation mechanism 15, the position / size detection mechanism 17, and the domain normalization mechanism 27. Illustrate the visual device from the 6th to the 10th patent application. In order to memorize the digital image in, the image memory mechanism 12 (refer to Fig. 6 and Fig. 6) implemented by the data processing device 110 operates in a parallel array operation unit 40 in a grid pattern. Assume that the array operation unit 40 arranged in the “one row” column on the grid is ΑΟΙΙϋ, and the AOIL algorithm is as shown in FIG. 14. In step 1201, AoUu is arranged in a row j column of the grid. Whether logically or physically, this is to determine what is needed near A0UU. 76 This paper size is applicable to China National Standard (CNS) A4 (210 X 297 mm) --------------- ^ Packing -------- Order ---- ----- line «C Please read the notes on the back before filling out this page) Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 510010 A7 B7 V. Description of the invention (7?) In step 1202, set the AOUu near and The initial ramp of the variable. In step 1203, it is determined whether or not the sequentially input digital images 111 are no longer present. If there is no digital image H1 (step 1203: YES), the algorithm ends. If there is a digital image 111 (step 1203: NO), it moves to step 1204. However, when assembling the array operation unit 40 only for a specific image size, even an infinite loop may be used. In step 1204, until the preparation of the digital image 111 is completed, an input wait is performed. In step 1205, the frequency band number portion of the pixels in row i and column j of the digital image 1II is input. Therefore, the AOUu needs at least a memory 42 for storing image data in the number of frequency bands. In step 1206, in order to enable output during the input waiting period, pixels in row 1 and column j of the billion-bit digital image 111 are recorded. In step 1207, a band pixel 値 of the digital image 111 is output. After that, it returns to step 1203. Accordingly, by using the data processing device 110 constituted by the array operation unit 40, the visual device corresponding to the image memory mechanism 12 can store the digital image 111. Array operation unit arranged in a grid pattern to vibrate the digital image 111 in order to vibrate the digital image 111 using the image vibration mechanism 13 (Patent 3, 4 and 5) of the patent application scope item 6 realized by the data processing device 110 The 40 series moves in parallel. It is assumed that the array operation unit 40 arranged in a grid form is AOUu, and the algorithm of the frame O AuU is shown in FIG. 15. In step 1301, ΑΙΙΙϋ is arranged in rows i and j of the grid. 77 The size of this paper is applicable to China National Standard (CNS) A4 (210 X 297 mm) ------------- Installation --------- Order ------ --- line (please read the notes on the back before filling this page) 540010 A7 B7 V. Description of the invention (Ί ") No matter whether it is logical or physical, this is to determine the need for Aoui" nearby. (Please read the notes on the back before filling this page.) In step 1302, set the vicinity of AOUu and the initial value of the variable. In step 1303, it is determined whether or not the sequentially input digital image Π1 is no longer present. If there is no digital image 111 (step 1303 ... YES), the algorithm ends. If there is a digital image 111 (step 1303: NO), the process moves to step 1304. However, when the array operation unit 40 is performed only on a specific image size, even an infinite loop may be used. In step 1304, the frequency band number portion of the pixels in row i and column j of the digital image 111 is input. Therefore, AOUu needs at least a memory 42 of image data in the number of hundreds of millions of frequency bands. In step 1305, according to the function Ξ # (χ), the image of row 1 and column J of image 111 is moved to one pixel of the nearby image. In step 1306, a band pixel 値 of the digital image 111 is output. After that, it returns to step 1303. According to this, using the data processing device 110 composed of the array operation unit 40 and the vision device corresponding to the 6th patent application scope of the image memory mechanism 13, the digital image 111 can be vibrated. The Consumer Information Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs printed the edge information generation mechanism 14 (refer to Fig. 1 and Fig. 6) of the patent application scope implemented by the data processing device 110 to generate rough edges from the digital image 111 The information image 113 and the array operation unit 40 arranged in a grid are synchronized and operated in parallel. Assuming that the array operation unit 40 arranged in the "i row" column on the grid is AOUu, the A0Uu algorithm corresponding to the edge information generating mechanism 14 is shown in FIG. 78 This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm) 540010 Printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7___ V. Description of the invention (Ή) In step 1401, arrange ΑΙΙΙϋ in the grid On row i and column j. Whether logically or physically, this is to determine what is needed near A0UU. In step 1402, the vicinity of AOUu and the initial value of the variable are set. In the neighborhood setting, the vicinity size q used by the aforementioned functions may be individually determined to be 4 or 8, or all of them may be unified to 4 or 8. In order to improve the accuracy of the generated rough edge information 112, the edge information generation mechanism 14 of the present invention controls the calculation time and the number of frequency bands of the digital image 111, etc. The edge information generation mechanism 14 may change the appropriate nearby size to deal with it as necessary O In step 1403, it is determined whether the digital image 111 has ended. If there is no digital image 111 (step 1403: YES), the algorithm ends. If there is a digital image 111 (step 1403: NO), the algorithm is also terminated. However, when the array operation unit 40 is assembled for a specific number of frequency bands and image sizes, even an infinite loop may be used. In step 1404, the frequency band number portion of the pixels in row i and column j of the digital image 111 is input. This is because AOlLj processes the relationship of pixels in one row and one column of the digital image in at a time. Therefore, A0 仏 needs at least a memory 42 for storing image data of a number of frequency bands. In step 1405, Aouu uses the communication with the nearby array operation unit 40 to smooth the pixels 値 of each frequency band of the input digital image 111 according to Sm (x). The smoothed band pixels 値 are treated as band pixels 値 of the smoothed image. Here, the function Sm (x) can also be repeated as many times as necessary. For general color images, twice is sufficient. 79 This paper size applies to China National Standard (CNS) A4 specification (210 X 297 mm) " 一 ------------ Installation -------- Order ---- ----- Line ΦΓ (Please read the notes on the back before filling out this page) 540010 Printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7 V. Description of the invention (y) In step 1406, Band pixels 値 are logarithmically transformed according to the function Luk (x). Band pixels that have undergone logarithmic conversion are treated as band pixels for gastric logarithmic conversion images. In step 1407, the AOUu uses the communication with the nearby array operation unit 40 to logarithmically transform each band pixel 値 in accordance with E # 00 to clarify. The sharpened band pixels are treated as band pixels of the sharpened image. In step 1408, each band pixel 値 'of the sharpened image is subtracted from each band pixel 値 of the sharpened image before input 1 according to the function DWx, y). The band pixel 値 for which the difference is calculated is processed as a band pixel 时间 of the time difference image. In step 1409, pixels in each frequency band corresponding to the sharpened image are replaced with pixels in each frequency band of the sharpened image before input 1. In step 1410, AOUu uses the communication with the nearby array operation unit 40 to calculate the Laplacian operation code with X based on the operator ▽ 2 for the band pixels 时间 of the time difference image. The band pixel 値 of the Laplacian opcode is calculated as the band pixel 値 of the time-differential Laplacian opcode image. In step 1411, AOlLj uses the communication with the nearby array operation unit 40 to extract the zero point for each band pixel 値 of the time difference Laplacian opcode image according to the function Z4χ). The band pixels 値 from which the zero point is extracted are treated as band pixels 时间 of the time difference zero point image. In step 1412, for each band pixel 値 of the time difference Laplacian opcode image, a maximum of 値 80 is detected in each band pixel according to the function Bm (x). 210 X 297 mm) " ------------ install -------- order --------- line (please read the precautions on the back before filling (This page) 540010 A7 B7 printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 5. Description of Invention (1). The detected maximum "band pixel" is treated as the maximum "band time pixel" of the time difference zero point image. For convenience, the number of frequency bands is one. In step 1413, AOUu uses the communication with the nearby array operation unit 40 to calculate the Laplacian operation code for each band pixel of the sharpened image based on the operator. The band pixel 値 of the Laplacian opcode is calculated as the band pixel 値 of the time-differential Laplacian opcode image. In step 1414, AOUu uses the communication with the nearby array operation unit 40 to extract the zero point of each band pixel 値 of the Laplacian code image according to the function Zμ (χ). The band pixels 値 from which the zero point is extracted are treated as band pixels 时间 of the time difference image. In step 1415, for each band pixel 値 of the Laplace opcode image, a maximum 値 is detected in each band pixel according to a function BuKx). The detected maximum "band pixel" is used as the maximum "band time pixel" of the time difference zero point image. For convenience, the number of frequency bands is one. In step 1416, each band pixel 値 of the Laplacian code image and each band pixel 时间 of the time difference Laplacian image are located in each image according to the function M # (x, y) Among the band pixels 値 at the same position, the maximum 値 is detected. The detected maximum band pixel 値 is planted as a band pixel of the mixed zero point image. For convenience, the number of frequency bands is one. In step 1417, AOUu uses the communication with the nearby array operation unit 40 to remove the holes for the band pixels 混合 of the mixed zero point image according to the function F # (x). The band pixel 値 after removing the hole is used to remove the mixed zeros. 81 This paper size applies to the Chinese National Standard (CNS) A4 specification (210 X 297 mm). ----- Order --------- Line · (Please read the precautions on the back before filling this page) 540010 Printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7 V. Invention Description (p) The band pixels of the image are 値. For convenience, the number of frequency bands is one. Here, the function Fm (x) can be repeated several times as needed. For general color images, once is sufficient. In step 1418, the person 011 " remove the out-of-band pixels and out-of-holes based on the function ΑυφΟ for the band pixels 値 of the mixed-zero image of the removal holes by using communication with the nearby array operation unit 40. Band pixels 値 after removing isolated points and holes are treated as band pixels 除去 for removing noise mixed zero-point images. For convenience, the number of frequency bands is one. In step 1419, for the band pixels 値 from which the mixed zero-point image is removed, 0 and 1 are inverted according to the function 1 μ (χ). The inverted band pixel 値 is processed as a band pixel 粗 of the rough edge information image 113. In step 1402, a band pixel 値 of the rough edge information image 113 is output. After that, the process returns to step 1403. Therefore, using the data processing device 110 constituted by the array operation unit 40 and corresponding to the visual device of the 7th patent application scope of the edge information generating mechanism 14, the rough edge information image 113 can be generated from the digital image 111. As shown in FIG. 17, the edge information formation mechanism 15 implemented by the data processing device 110 (refer to FIGS. 1 to 5 and 7 to 12), in order to apply the eighth of The rough edge information image 113 and the digital image 111 generate a formation edge information image 115 composed of the formation edge information 114, and the array operation unit 40 arranged in a grid pattern operates in parallel and in parallel. Let the array operation unit 40 arranged in the "row" on the grid be AOUu, and the algorithm of AoUu is shown in Fig. 18. In step 1501, AoUu is arranged in row i and column j. Regardless of the size of the logical 82 paper, the Chinese National Standard (CNS) A4 specification (210 X 297 mm) is applicable --- f --- 〇 ------ installation --------- order-- ------- Line (Please read the notes on the back before filling this page) 540010 A7 B7 V. Description of the Invention (yl) Compilation or physical, this is to determine the need for AOUU nearby. (Please read the notes on the back before filling this page.) In step 1502, set the vicinity of AOUu and the initial value of the variable. When setting the neighborhood, the neighborhood size q used by each of the aforementioned functions can be individually determined to be 4 or 8.. Alternatively, all of them can be unified to 4 or 8. In order to improve the accuracy of the formation of the edge information 114 formed by the edge information information formation mechanism 15 of the present invention, it is better to set all nearby sizes q to 8. However, in order to form the rough edge information 112, the edge information forming mechanism 15 may appropriately change the nearby size to handle the situation, depending on the limitation of the calculation time and the number of frequency bands of the input digital image 111. In step 1503, it is determined whether the sequentially input digital image 111 or the rough edge information image 113 is no longer present. If neither the digital image 111 nor the rough edge information image 113 is present (step 1503: YES), the algorithm is terminated. If both the digital image 111 and the rough edge information image 113 are present (step 1503: NO), the process proceeds to step 1504. However, when the array operation unit 40 is assembled for a specific number of frequency bands and image size, it may be used as an infinite loop. Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. In step 1504, the frequency band portion of the pixels in the “i-th” row of the digital image 111 and the rough edge information image 113 is input. This is because AOUu processes the relationship between the pixels in the i-th row and the j-th column of the digital image 111 and the rough edge information image 113 at a time. Therefore, the AOIL needs at least a memory 42 for storing image data in a number of bands. In step 1505, pixels in one row and column of the digital image 111 are separated from pixels in row i and column j of the rough edge information image 113. This is because A〇Uu processes the relationship between the pixels of one row of the digital image 111 and the pixels of the row j of the rough edge information image 113 as separate image pixels. 83 $ Paper scale is applicable to China National Standard (CNS) A4 Specification (210 X 297 Public Love) Printed by the Consumers' Cooperative of the Intellectual Property Bureau of the Ministry of Economy 540010 A7 B7 V. Description of Invention (^ JL). If the i-th: i-th pixel of the digital image 111 and the i-th j-th pixel of the rough edge information image 113 are input in the separated state at the beginning, no processing is required. In step 1506, AOUu uses the communication with the nearby array operation unit 40 to smooth the pixels 値 of each frequency band of the input digital image 111 according to the function Suk (x). The smoothed band pixels 値 are treated as band pixels 値 of the smoothed image. Here, the function SUk (x) can also be repeated as many times as necessary. For general color images, twice is sufficient. In step 1507, the pixels in each band of the smoothed image are log-transformed according to Luk (x). The band-converted pixel 値 of the logarithmic conversion is treated as the band-pixel 値 of the log-transformed image. In step 1508, AOlLj uses the communication with the nearby array operation unit 40 to refine the pixel 値 according to the function Eϋφ0. The sharpened band pixels are treated as band pixels of the sharpened image. · In step 1509, ΑΙΙΙ is a Laplacian operation code calculation of pixels in each band of the sharpened image using communication with a nearby array operation unit 40, according to operation v2ljkX. The band pixel 値 of the Laplace opcode is calculated as a band pixel 图像 of the Laplace opcode image. In step 1510, AOUu uses the communication with the nearby array operation unit 40 to extract the zero point of each band pixel 値 of the Laplace opcode image according to the function Z ^ (x). The band pixels 値 of the zero point are extracted and processed as band pixels 値 of the zero image. 84 This paper size is applicable to China National Standard (CNS) A4 specification (210 X 297 mm) ---.--- f I ---- Loading -------- Order ------- --Line 1 ^ 1 " (Please read the notes on the back before filling out this page) Printed by the Employees' Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 540010 A7 B7 V. Description of the invention (浐 y In step 1511, the zero point image The maximum pixel 値 is detected in each band pixel 根据 according to the function B ^ U). The detected maximum band pixel 値 is processed as a band pixel 値 of the maximum 値 zero point image. The number of frequency bands is 1 for convenience. In step 1512, for the band pixels 値 of the maximum 値 zero point image, 0 and 1 are inverted according to the function 1 ^ 00. The inverted band pixels 値 are processed as band pixels 图像 of the basic edge information image. In step 1513, the input band edge pixel 粗 of the rough edge information image 113 is initially processed as the band pixel 整 of the shaped rough edge information image. The band pixel 値 of the information image is shaped by the function Qw (x, y) of the band pixel 整 of the shaped rough edge information image. The shaped band pixels 値 are processed as band pixels 整 of the shaped rough edge information image. Here, the function Q4x, y) is repeated until the band pixels 値 of the information image of the rough-shaping original shape are no longer changed. However, according to the limitation of the calculation time ^, the quality of the input rough edge information image 113, the quality required to form the formed edge information image 115, etc., it is better that the shaping process ends the algorithm by an appropriate number of iterations. In step 1514, AOUu uses the communication with the nearby array computing unit 40 to perform a line width interpolation on the band pixels 整 of the shaped rough edge information image according to the function Cuk (x). The interpolated band pixels 値 are processed as band pixels 形 forming the edge information image 115. In step 1515, a band pixel 値 forming the edge information image II5 is output. After that, the process returns to step 1503. 85 --i --- U ------ install -------- order --------- line ^ ΙΓ (Please read the precautions on the back before filling this page) This The paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm) 540010 A7 ____ B7 V. Description of the invention (#) Therefore, the data processing device 110 composed of the array operation unit 40 is used, corresponding to the edge information forming mechanism The visual device of the eighth patent application range of 15 can generate a formation edge information image 115 composed of a rough edge information image 113. Here, the formation from the rough edge information image 113 to the formation edge information image 115 can be regarded as the edge information generated from shooting a low-resolution digital image 111 of a certain situation, and the high-resolution digits of the same situation are estimated. Edge information generated by the image 111. Here, for the natural number η, as shown in FIG. 19, when the low-resolution coarse edge information image 117 is generated using the edge generation mechanism η based on the low-resolution digital image 116 that reduces the resolution i / η of the digital image U1, the The low-resolution coarse-edge information image U7 is enlarged by η times to generate a coarse-edge information image ία. However, in order to set the resolution of the digital image 111 to 1 / η, one of the n consecutive digital images 111 may be simply extracted in the horizontal and vertical directions. In addition, in order to enlarge the low-resolution coarse-edge information image 117 by η times, the pixels of the continuous inter-band frequency band of the low-resolution coarse-edge information image 117 of the low-resolution coarse-edge information image 117 can be made up of η-1 in the horizontal and vertical directions. . At this time, if η is not too large, the data processing device 11 of the edge information forming mechanism 15 is realized to form an edge information image 115 that forms a coarse edge information image 113 that enlarges the low-resolution coarse edge information image 117, and implements The data processing device 110 of the edge information forming mechanism 15 forms the rough edge information image 115 of the rough edge information image 113 generated from the digital image 111, which is substantially the same. The reason for this is because the edge information forming mechanism 15 uses which edge information is used as a reference in the edge information generated internally by using the digital image 111. 86 * This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm) (Please read the precautions on the back before filling out this page) Order --------- Printed by the Consumer Consumption Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs Printed by the Employee Consumption Cooperative of the Intellectual Property Bureau of the Ministry of Economy 540010 A7 ________ B7 V. Explanation of the invention The edge information forming mechanism 15 only uses the relationship of the coarse edge information image 113. Therefore, when the coarse edge information image 113 of the low-resolution coarse edge information image 117 is enlarged and input to the edge information forming mechanism 15, the edge information generation mechanism 13 for generating the coarse edge information image in from the low-resolution digital image 116 is implemented. The data processing device 11 can reduce the amount of hardware. If this method is further developed, as shown in FIG. 20, the rough edge information 112 around the low-resolution coarse edge information image 117 generated by the low-resolution digital image 116 that reduces the resolution of the digital image 111 can be generated. The low resolution cuts out the rough edge information image 118. If the low-resolution coarse-edge information image 118 is enlarged and the coarse-edge information image 119 is cut out, the digital image 120 in the same area as the digital image 111 is cut out, and the edge information forming mechanism 14 is input to implement If the data processing device 110 is used, the edge information image 121 can be generated by cutting out. At this time, the data processing device 11 of the edge information forming mechanism 14 is realized, and the amount of hardware can be reduced. As shown in FIG. 21, the position / size detection mechanism 17 of the patent application scope item 9 implemented by the data processing device 110 (refer to FIG. 1 and FIG. 6) is to change the rough edge information 112 as the thickness of the pixels. The edge information image 113 generates a repeated information image 132 having the repeated information 131 as pixels, and the array operation unit 40 arranged in a grid pattern operates in parallel and in parallel. Assuming that the array operation unit 40 arranged in the i row and the j column on the grid is AoUu, the algorithm of AOU ′ is shown in FIG. 22. In step 1701, AOU, "is arranged on the grid in a row j 歹 fj. This is theoretical or physical. This is to determine what is needed near A0UU. In step 1702, the vicinity of AOUu and the initial value of the variable are set. 87 This paper size applies to China National Standard (CNS) A4 specification (210x 297 mm) --i --------- installation -------- order -------- line (Please read the precautions on the back before filling this page) 540010 Printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7 V. Description of the invention (^) When setting nearby, you can individually set the nearby size used by each of the aforementioned functions q, can all be unified. In order to improve the accuracy of the generated duplicate information image 132, the data processing device 110 according to the present invention preferably sets all the adjacent sizes q to a larger value. However, if the position / size detection mechanism Π can appropriately change the nearby size if necessary due to the limitation of the calculation time used to calculate the center of gravity of the rough edge information 112 of the object, the size of the input rough edge information image 113, etc. . In step 1703, it is determined whether the sequentially input rough edge information image 113 is no longer present. If there is no rough edge information image 1Π (step Π03: YES), the algorithm ends. If there is a rough edge information image 113 (step 1703: NO), it moves to step 1704. However, when the array operation unit 40 is assembled only for a specific image size, it may be used as an infinite circuit. In step 1704, one row of the rough edge information image 113 is input: the pixel 1 band portion of the i column. Therefore, AOlLj needs at least a memory 42 for storing image data in a frequency band. In step 1705, the rough edge information 112 of the rough edge information image 113 is converted into the repeat information 131 of the repeat information image 132. The duplicate information 131 is a band pixel 値 corresponding to 1 or 0. In step 1706, A0U. The communication amount with the nearby array operation unit 40 is used to calculate the movement amount of each band pixel 値 of the repeated information image 132 based on the function ΔμΟΟ. The band pixel 値 that has been imaged with the movement amount is treated as the band pixel 値 of the movement amount image. In step 1707, AOUu uses the communication with the nearby array operation unit 40 to make each frequency band of the repeating information image 132 according to the function Λ, KK). This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297). (Mm): U ------------— ^ --------- (Please read the notes on the back before filling this page) 540010 A7 B7 V. Description of the invention ( According to this, using the data processing device 110 composed of the array operation unit 40 and the visual device corresponding to item 9 of the patent application range of the position / size detection mechanism 17 can generate the repeated information image 132 from the rough edge information image 113 . Of course, the visual device of the ninth patent application scope can also generate the repeated information image 132 from the edge information image 115 without using the rough edge information image 113. Therefore, using the data processing device 110 constituted by the array operation unit 40 and the vision device corresponding to the position / size detection mechanism 17 can generate the repeated information image 132 from the edge information image 115. In the algorithm of FIG. 22, if the object area image 142 that displays the object area 141 is used instead of the rough edge information image 113 composed of the rough edge information 112, as shown in FIG. 23, the data processing device 110 is used. The implemented position / size detection mechanism 17 (see FIGS. 2 and 4) can also generate a repeat information image 132 displaying the repeat information 131 from the object area image 142 displaying the object area 141. However, when the object area image 142 is used, the repeated information image 132 of the repeated information 131 is displayed based on the total number of pixels of the object area 141 centered on its position. Therefore, the result means an object centered on its position. Area. Therefore, when obtaining the size of an object from the repeated information image 132, it is necessary to pay attention to using the square root of the repeated information ι31 ° As shown in FIG. 24, the field standardization mechanism of the tenth patent application range realized by the data processing device 27 (refer to FIG. 7), in order to generate an object area image 142 including an object area 141 and a normalized image 145 including a specification area 144 from a digital image 111, the grid computing unit 40 is performed in parallel in parallel. action. Set a row of 90 on the grid. This paper size is applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) (Please read the precautions on the back before filling this page) ▼ Installation ------- Order --------- Printed by the Consumer Property Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs, printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economics, printed by 540010 A7 B7 V. Description of the Invention (θ) The array operation unit 40 in column j is ΑΟΙΙϋ, the algorithm of AOUu is shown in Figure 25. In step 2701, AOUu is arranged on the grid in i rows and columns. This is theoretical or physical. This is to determine what is needed near AOUij. In step 2702, the vicinity of AOU, and the initial value of the variable are set. When setting the neighborhood, the neighborhood size q used by the aforementioned functions can be individually set, or all of them can be unified. In the field normalization mechanism 27 of the present invention, in order to improve the accuracy of the generated normalized image 145, it is preferable to set all the nearby sizes Q to a large value. However, in order to limit the calculation time used for normalizing the separated object domain 143, the size of the input digital image 111, and other factors, the domain normalization mechanism 27 may appropriately change the nearby dimensions as necessary to perform the disposal. In step 2703, it is determined whether or not the object area image 142 or the digital image 111 that have been sequentially input are no longer present. If there is no object area image 142 or digital image 111 (step 2703: YES), the algorithm ends. If there is an object area image 142 or a digital image 111 (step 2703: NO), it moves to 2704. However, when the array operation unit 40 is assembled only for a specific number of frequency bands and image sizes, it may be set to an infinite loop. In step 2704, pixels in the i-th column of the object area image 142 in the 1-band portion and pixels in the 1-th row and j-column of the digital image 111 in the frequency band portion are input. This is because A0% deals with the relationship between the pixels in row i and column j of the object area image 142 and the pixels in row j and column 1 of the digital image 111 at one time. Therefore, the AOUu needs at least a memory 42 for memorizing image data of the total number of frequency bands. 91-· ---: ------ Install -------- Order --------- (Please read the precautions on the back before filling this page) This paper size applies China National Standard (CNS) A4 specification (210 X 297 mm) 540010
經濟部智慧財產局員工消費合作社印製 五、發明說明（y) , 在步驟2705中，分離物體領域圖像I42的i行j列的 像素與數位圖像111的i行」列的像素。這是因爲A0仏將 物體領域圖像142的1行j列的像素與數位圖像111的1行 j列的像素作爲各自獨立的圖像像素來處理的關係。若物體 領域圖像142的i行j列的像素與數位圖像111的1行」列 的像素，從一開始就被分離輸入的話，無需進行任何處理 。物體領域圖像142與數位圖像111 ’分別被複製在更新 物體領域圖像及數位圖像中。 在步驟2706中，AOUu是利用與附近的陣列運算單元 40之通信，對更新物體領域圖像的各頻帶像素値’根據函 數Rui(x)計算移動量。已將移動量圖像化的頻帶像素値， 被作爲移動量圖像的頻帶像素値加以處理。 在步驟2707中，AOUu是利用與附近的陣列運算單元 40之通信，可根據函數Hm(x，y)自更新物體領域圖像的各 頻帶像素値中找到能移動的移動端頻帶像素値。顯示是否 爲能移動之移動端的値，被作爲能移動圖像的頻帶像素値 加以處理。 在步驟2708中，AOUu是利用與附近的陣列運算單元 40之通信，根據函數Uljk(x，y)將更新物體領域圖像的各頻 帶像素値移動到能移動端。移動的頻帶像素値，重新被作 爲新物體領域圖像的各頻帶像素値加以處理。 在步驟2709中，AOUu是利用與附近的陣列運算單元 4〇之通信，根據函數，y)將更新物體領域圖像的各頻 帶像素値移動到能移動端。移動的頻帶像素値，重新被作 92 --*---：------裝--------訂---------線 (請先閱讀背面之注意事項再填寫本頁) 本紙張尺度適用中國國家標準（CNS)A4規格(210 X 297公釐） 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明/) 爲新物體領域圖像的各頻帶像素値加以處理。 在步驟2710中，判斷顯示從步驟2706到步驟2709的 反覆次數的移動次數，是否已達到指定次數。若移動次數 未達指定次數的話(步驟2710 : NO)，即回到步驟2706。若 移動次數達到指定次數的話(步驟2710 : YES)，即移到步驟 2711。又，該指定次數是由數位圖像1U的大小與數位圖 像111的分離物體領域143的大小，及附近的大小q來決 定。若視利用目的設定適當之參數的話，即使大致決定指 定次數亦無所謂，但若指定次數過多的話，規格化所需要 的時間就會變長。 在步驟2711中，AOUu是利用與附近的陣列運算單元 40之通信，對完成移動的更新物體領域圖像的各頻帶像素 値，根據函數V#(x ’ y)用附近的平均値進行內插。又，X 與y同時成爲更新物體領域圖像。用平均値彌補的頻帶像 素値，被作爲規格化的更新物體領域圖像的各頻帶像素値 加以處理。 在步驟2712中，AOUu是利用與附近的陣列運算單元 40之通信，對完成移動的更新數位圖像的各頻帶像素値， 根據函數Vl|k(x ’ y)用附近的平均値進行內插。又，X爲更 新數位圖像，y爲更新物體領域圖像。用平均値彌補的頻 帶像素値，被作爲規格化的更新數位圖像的頻帶像素値加 以處理。 在步驟2713中，判斷顯示從步驟2711到步驟2712的 反覆次數的內插次數’是否已達到指定次數。若內插次數 93 ^紙張尺度適用中國國家標準(CNS)A4規格（210 X 297公釐） ---·---^------·11111--^ --------I C請先閱讀背面之注音？事項再填寫本頁} 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明/ 未達到指定次數的話(步驟2713: N0)，即回到步驟2711。 若內插次數達到指定次數的話(步驟2713: YES)，即移到步 驟2714。一般的內插次數只要附近大小q的一半左右次數 ，即已足夠。 在步驟2714中’判斷顯不從步驟2706到步驟2713的 反覆次數的持續次數，是否已達到指定次數。若持續次數 未達到指定次數的話(步驟2714 : NO)，即回到步驟2706。 若持續次數達到指定次數的話(步驟2714 : YES)，即移到步 驟2715。又，該指定次數是由數位圖像111的大小與數位 圖像111的分離物體領域143的大小’及附近的大小q來 決定。若視利用目的設定適當之參數的話，即使大致決定 指定次數亦無所謂，但若指定次數過多的話，規格化所需 要的時間就會變長。 在步驟2715中，更新數位圖像的頻帶像素値是以規格 化圖像145的頻帶像素値加以輸出。之後，回到步驟2703 〇 因此，使用由陣列運算單元40所構成的資料處理裝置 Π0，對應領域規格化機構27的申請專利範圍第10項之視 覺裝置，能從物體領域圖像142及數位圖像，生成規格化 圖像145。 利用資料處理裝置110所實現的規格化圖像保持機構 28(參照第7圖），爲了記憶規格化圖像145，網格狀排列的 陣列運算單元40是同步並列進行動作。設格子上1行」列 所配置的陣列運算單元40爲AOU,，則AOUu的算法如第 94 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） —,—·------^ —------訂--------- (請先閱讀背面之注意事項再填寫本頁) 540010 經濟部智慧財產局員工消費合作社印制衣 A7 B7 五、發明說明（丫)） 〆 26圖所示。 在步驟2801中，將AOUu配置在格子上！行列上。 無論在邏輯上或物理上，此係爲了決定AOUij的附近所需 者。 在步驟2802中，設定AOUu的附近與變數的初始値。 在步驟2803中，判斷依序輸入的規格化圖像145是否 已經沒有。若無規格化圖像145的話(步驟2803 : YES)，即 結束算法。若有規格化圖像145的話(步驟2803 : NO)，即 移到步驟2804。但，當僅對特定的圖像尺寸組裝陣列運算 單元40時，作爲無限回路亦可。 在步驟2804中，輸入規格化圖像145的i行」列像素 的頻帶數部分。因此，AOUu至少需要記憶頻帶數部分的圖 像資料的記憶體42。 在步驟2805中，若需要輸出端的裝置的話，即轉換規 格化圖像145的格式。特別是，當將規格化圖像丨45的頻 帶數設定爲1，或當數位圖像111的頻帶數爲4以上時將 規格化圖像145的頻帶數設定爲3，以使類比信號較易生 成時非常方便。否則〃不需進彳了任何處理。 在步驟2806中，爲了能確實地傳送圖像資料至處理速 度不同的輸出端裝置，記憶規格化圖像145的i行」列的 像素。 在步驟2807中，輸出規格化圖像145的頻帶像素値。 之後，回到步驟2803。 據此，使用由陣列運算單元40所構成的資料處理裝置 95 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） --,----------·裝--------訂--------- (請先閱讀背面之注意事項再填寫本頁) 540010 A7 B7_______ 五、發明說明（ Η ，對應規格化圖像保持機構2 8的視覺裝置，能輸出規格化 圖像145 〇 (請先閱讀背面之注意事項再填寫本頁) 上述已說明了圖像記憶機構12(參照第1圖及第6圖) 、圖像振動機構13(參照第3圖、第4圖及第5圖）、邊緣 資訊生成機構14(參照第1圖及第6圖）、邊緣資訊形成機 構15(參照第1〜5圖、第7〜12圖）、物體/背景分離機構 16(不參照第2、4、6、7〜12圖）、位置/大小檢測機構17( 參照第1圖及第6圖）、領域規格化機構27(參照第7圖）、 及規格化圖像保持機構28(參照第7圖）的算法。該等機構 能全部利用附近處理來實現，故該等機構能使用網格狀排 列的陣列運算單元40所構成的資料處理裝置ho。但，若 僅用圖像辨識機構29(參照第8圖)即不易實現。因此，在 圖像辨識機構29中，爲了檢討是否能利用網格狀排列的陣 列運算單元40所構成的資料處理裝置11〇，係對圖像辨識 機構29的處理加以詳細分類。 經濟部智慧財產局員工消費合作社印製 圖像辨識機構29，係對整個規格化圖像145(參照第24 圖）要求自事先準備的候補中特定出放大之移動物體2或靜 止物體3，以生成辨識結果。特定移動物體2或靜止物體3 的最基本方法，是儘量準備多數之移動物體2或靜止物體 3的模板圖像146(參照第27圖），藉比較規格化圖像145與 模板圖像146,即能找出最像規格化圖像145的模板圖像 146。然而，圖像辨識機構29，若僅從規格化圖像145及模 板圖像146中抽出任意的像素進行比較，則由於不能找出 最像規格化圖像145的模板圖像146，故必須比較規格化 96 冢紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） " 540010 五 ___ 經濟部智慧財產局員工消費合作社印製 A7 B7 發明說明（ 圖像145及模板圖像146的所有像素。因此，圖像辨識機 構29需要用最小平方法與中性網路等的大區域處理。由於 資料處理裝置110是適合附近處理的構造，故僅使用資料 處理裝置110，不易實現圖像辨識機構29。 然而，圖像辨識機構29，不需要在從規格化圖像145 到生成辨識結果的整徹過程中進行大區域處理。亦即，從 規格化圖像145及模板圖像146中抽出任意的像素，根據 比較的結果生成辨識結果的過程中，圖像辨識機構29雖需 要大區域處理，但從規格化圖像145及模板圖像146抽出 任意的像素加以比較的過程中，圖像辨識機構29未必需要 大區域處理。此處，由於從規格化圖像145及模板圖像 146中抽出任意的像素加以比較的過程，係最基本的圖形 匹配，因此若該圖形匹配能利用附近處理加以實現的話， 即能以執行多數決定等簡單的數値計算之一般處理器’僅 實現根據此圖形匹配結果生成辨識結果的過程。因此’以 下就使用資料處理裝置來實現圖形匹配的方法’加以 說明。 首先，設規格化圖像145爲X，η個模板圖像146爲y1 、γ2···/···γη。當使用自然數g時，匹配結果圖像丨47的i 行」列的匹配結果，是根據式37比較規格化圖像145 及模板圖像146的1行j列的像素，指出具有最像規格化圖 像145像素的模板圖像146的號碼。又’因匹配結果圖像 147爲單頻帶圖像，故爲方便起見，以頻帶數1的圖像進 行處理。因此，匹配結果5,"的第3添加字爲1。 97 本紙張尺(CNS)A4^ (210 x 297 ^ ) —.—一------^--------^---------^9. (請先閱讀背面之注意事項再填寫本頁) 540010 A7 B7 經濟部智慧財產局員工消費合作社印制π 五、發明說明（% 9 if - vijk)2 ^ min^A^n Eki^ijk - y^k)： %i — \ ίοτ 1 < g <n and only one of g, 0 otherwise. (37) 此處，根據式37所生成的匹配結果5ϋΐ在整個匹酉己糸吉 果圖像147中，未必統一。當有許多模板圖像146日寺，匹 配結果圖像147成爲馬賽克狀的可能性很高。因此，胃米斗 處理裝置110即計算匹配結果與相對該Q附近內之匹 配結果的直方圖，將集中匹配結果的方法顯示如下。 當任意的單頻帶圖像X是匹配結果圖像147時，若使 用自然數g、實數11與〃的話，匹配圖像172即根據式38 及39進行更新。又，匹配結果圖像147因爲單頻帶圖像， 爲方便起見，以頻帶數1的圖像進行處理。因此，函數Ψ ϋΐ(Χ)的第3添加字爲1。 φΰ·1(χ) 9 辽 eq(g，〜〇 + Σ吣，m，i)€iV(?)叫(5, Ami)= {eq(/i, + ΣΖρ(/,Γη,ΐ)€Ρ{；·ι((7) and 2{eq{g)xijl) + Ep(/,m,i)GPl;i((?) e(l(5^ ^ ^i；i for 1 < g < n and only one of g, 0 otherwise. eq(乜，Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 5. Description of the invention (y), in step 2705, the pixels in row i and column j of the object area image I42 and the pixels in row i and column of the digital image 111 are separated. This is because A0 仏 treats the pixels in row 1 and column J of object area image 142 and the pixels in row 1 and column j of digital image 111 as independent image pixels. If the pixels in row i and column j of the object area image 142 and the pixels in row one and column digit of the digital image 111 are separated and inputted from the beginning, no processing is required. The object area image 142 and the digital image 111 'are copied in the updated object area image and the digital image, respectively. In step 2706, AOUu uses the communication with the nearby array operation unit 40 to calculate the amount of movement for each band pixel 値 'of the image of the object area to be updated based on the function Rui (x). The band pixel 値 that has been imaged with the movement amount is treated as the band pixel 値 of the movement amount image. In step 2707, AOUu uses the communication with the nearby array operation unit 40 to find a mobile band pixel 値 that can be moved from each band pixel 更新 of the object area image update according to the function Hm (x, y). Whether or not the display is a mobile mobile terminal is treated as a band pixel of the mobile image. In step 2708, AOUu uses the communication with the nearby array operation unit 40 to move each frequency band pixel 更新 of the object area image to the mobile end according to the function Uljk (x, y). The moving band pixels 値 are processed again as band pixels 图像 of the new object area image. In step 2709, AOUu uses the communication with the nearby array operation unit 40, and according to the function, y) moves each frequency band pixel 更新 of the object area image to the mobile end. The moving band pixels 値 are re-made as 92-* ---: ------ install -------- order --------- line (please read the note on the back first) Please fill in this page again for this matter) This paper size is applicable to China National Standard (CNS) A4 specification (210 X 297 mm) 540010 Printed by A7 B7, Consumer Cooperative of Intellectual Property Bureau of the Ministry of Economy The pixels of each frequency band are processed. In step 2710, it is judged whether or not the number of movements showing the number of iterations from step 2706 to step 2709 has reached the designated number. If the number of movements has not reached the specified number (step 2710: NO), the process returns to step 2706. If the number of moves reaches the specified number (step 2710: YES), the process moves to step 2711. The specified number of times is determined by the size of the digital image 1U, the size of the separated object area 143 of the digital image 111, and the nearby size q. If appropriate parameters are set depending on the purpose of use, it does not matter even if the number of designated times is roughly determined, but if the number of designated times is too large, the time required for standardization becomes longer. In step 2711, AOUu uses the communication with the nearby array operation unit 40 to interpolate the pixels 各 of each frequency band of the updated object area image that has been moved, and uses the nearby average 値 to perform interpolation according to the function V # (x 'y) . In addition, X and y simultaneously update the image of the object domain. The band pixels 値 compensated by the average 値 are processed as normalized pixels 各 in each band of the updated object area image. In step 2712, AOUu uses the communication with the nearby array operation unit 40 to interpolate the pixels 各 of each band of the updated digital image that has been moved, and uses the nearby average 附近 to perform interpolation according to the function Vl | k (x 'y). . In addition, X is an updated digital image, and y is an updated object area image. Band pixels 値 compensated by average 値 are treated as normalized band pixels 値 of updated digital images. In step 2713, it is determined whether or not the number of interpolations', which indicates the number of iterations from step 2711 to step 2712, has reached the specified number. If the number of interpolations is 93 ^ The paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm) --- · --- ^ ------ 11111-^ ------- -IC Please read the Zhuyin on the back? Please fill out this page again} Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 540010 A7 B7 5. Description of the invention / If the specified number of times has not been reached (step 2713: N0), return to step 2711. If the number of interpolations reaches the specified number (step 2713: YES), the process moves to step 2714. The general number of interpolations is sufficient as long as it is about half of the nearby size q. In step 2714, it is judged whether or not the number of times of repeating from step 2706 to step 2713 has reached the specified number of times. If the number of times has not reached the specified number (step 2714: NO), the process returns to step 2706. If the number of continuations reaches the specified number (step 2714: YES), the process moves to step 2715. The specified number of times is determined by the size of the digital image 111, the size of the separated object area 143 of the digital image 111, and the size q in the vicinity. If appropriate parameters are set depending on the purpose of use, it does not matter if the number of designated times is roughly determined. However, if the number of designated times is too large, the time required for standardization becomes longer. In step 2715, the band pixels 値 of the updated digital image are output as band pixels 规格 of the normalized image 145. After that, the process returns to step 2703. Therefore, using the data processing device Π0 composed of the array operation unit 40, the visual device corresponding to the 10th patent application scope of the field standardization mechanism 27 can obtain the image 142 and the digital image from the object field Image, a normalized image 145 is generated. In order to memorize the normalized image 145, the normalized image holding mechanism 28 implemented by the data processing device 110 (see FIG. 7) operates in a parallel array operation unit 40 in parallel. Suppose that the array operation unit 40 arranged in a row on the grid is AOU, and the algorithm of AOUu is the same as the 94th paper size. It applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm). ---- ^ ------- Order --------- (Please read the notes on the back before filling out this page) 540010 Employees' Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs printed clothing A7 B7 5 2. Description of the invention (Ya)) Figure 26 shows. In step 2801, AOUu is arranged on the grid! Ranks. Whether logically or physically, this is to decide what is needed near AOUij. In step 2802, the vicinity of AOUu and the initial value of the variable are set. In step 2803, it is determined whether or not the normalized image 145 that has been sequentially inputted is no longer present. If there is no normalized image 145 (step 2803: YES), the algorithm ends. If there is a normalized image 145 (step 2803: NO), the process moves to step 2804. However, when the array operation unit 40 is assembled only for a specific image size, it may be used as an infinite circuit. In step 2804, the frequency band portion of the pixels in the "i" row "column of the normalized image 145 is input. Therefore, AOUu needs at least a memory 42 for storing image data in the number of bands. In step 2805, if the output device is needed, the format of the normalized image 145 is converted. In particular, when the number of frequency bands of the normalized image 丨 45 is set to 1, or when the number of frequency bands of the digital image 111 is 4 or more, the number of frequency bands of the normalized image 145 is set to 3 to make analog signals easier Very convenient when generating. Otherwise, no further processing is required. In step 2806, in order to reliably transmit the image data to the output terminal devices having different processing speeds, the pixels in the "i" row and "column" of the normalized image 145 are stored. In step 2807, the band pixels 値 of the normalized image 145 are output. After that, the process returns to step 2803. Based on this, a data processing device composed of an array operation unit 40 is used. This paper size is in accordance with the Chinese National Standard (CNS) A4 specification (210 X 297 mm). -------- Order --------- (Please read the notes on the back before filling out this page) 540010 A7 B7_______ V. Description of the invention (Η, corresponding to the standardized image retention mechanism 2 8 The visual device can output a normalized image 145 〇 (Please read the precautions on the back before filling in this page) The image memory mechanism 12 (see Figures 1 and 6) and the image vibration mechanism 13 have been described above. (Refer to Figures 3, 4 and 5), edge information generating mechanism 14 (refer to Figures 1 and 6), edge information generating mechanism 15 (refer to Figures 1 to 5, and 7 to 12) Object / background separation mechanism 16 (not referring to Figures 2, 4, 6, 7 to 12), position / size detection mechanism 17 (refer to Figures 1 and 6), domain standardization mechanism 27 (refer to Figure 7) ), And the algorithm of the normalized image holding mechanism 28 (refer to FIG. 7). These mechanisms can all be implemented using nearby processing, so these mechanisms can use grid-shaped The data processing device ho constituted by the array operation unit 40. However, it is not easy to implement only with the image recognition mechanism 29 (see FIG. 8). Therefore, in the image recognition mechanism 29, in order to review whether the grid shape can be used The data processing device 11 constituted by the arrayed array computing unit 40 classifies the processing of the image recognition mechanism 29 in detail. The image recognition mechanism 29 printed by the staff consumer cooperative of the Intellectual Property Bureau of the Ministry of Economy prints the entire normalized map. Image 145 (Refer to Figure 24) requires the enlarged moving object 2 or stationary object 3 to be identified from the candidates prepared in advance to generate a recognition result. The most basic method for specifying moving object 2 or stationary object 3 is to prepare as many as possible. By comparing the template image 146 of the moving object 2 or the stationary object 3 (refer to FIG. 27) and comparing the normalized image 145 with the template image 146, the template image 146 that looks like the normalized image 145 can be found. However, If the image recognition mechanism 29 only extracts arbitrary pixels from the normalized image 145 and the template image 146 for comparison, the template image 146 that is most similar to the normalized image 145 cannot be found, so Must be compared to the normalized 96-gauge paper standard applicable to China National Standard (CNS) A4 (210 X 297 mm) " 540010 Five All pixels of the image 146. Therefore, the image recognition mechanism 29 needs to process a large area using a least square method and a neutral network. Since the data processing device 110 is a structure suitable for nearby processing, only the data processing device 110 is used. It is not easy to implement the image recognition mechanism 29. However, the image recognition mechanism 29 does not need to perform large-area processing during the entire process from the normalized image 145 to the generation of the recognition result. That is, in the process of extracting arbitrary pixels from the normalized image 145 and the template image 146 and generating a recognition result based on the comparison result, although the image recognition mechanism 29 needs to process a large area, the normalized image 145 and In the process of extracting arbitrary pixels for comparison in the template image 146, the image recognition mechanism 29 does not necessarily require large-area processing. Here, the process of extracting arbitrary pixels from the normalized image 145 and the template image 146 for comparison is the most basic pattern matching. Therefore, if the pattern matching can be realized by nearby processing, it can be executed by majority The general processor 'for simple data calculation such as decision' only implements the process of generating a recognition result based on the matching result of this figure. Therefore, "the method for realizing pattern matching using a data processing device" will be described below. First, let the normalized image 145 be X, and the n template images 146 be y1 and γ2... Γη. When the natural number g is used, the matching result of the "i-row" column of the matching result image 丨 47 "is a comparison of the pixels in row 1 and column j of the normalized image 145 and the template image 146 according to Equation 37, indicating that it has the most similar specifications. The number of the template image 146 of the image 145 pixels. Since the matching result image 147 is a single-band image, for convenience, an image with a frequency band number of 1 is used for processing. Therefore, the matching result 5, " is the third added word of 1. 97 paper rulers (CNS) A4 ^ (210 x 297 ^) —.— 一 ------ ^ -------- ^ --------- ^ 9. (Please first Read the notes on the back and fill in this page) 540010 A7 B7 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs π 5. Invention Description (% 9 if-vijk) 2 ^ min ^ A ^ n Eki ^ ijk-y ^ k) :% I — \ ίοτ 1 < g < n and only one of g, 0 otherwise. (37) Here, the matching result 5 generated according to Equation 37 is 5ϋΐ in the whole image 147, Not necessarily unified. When there are many template images 146 days, the matching result image 147 is highly likely to be mosaic-like. Therefore, the stomach rice bucket processing device 110 calculates a histogram of the matching result and the matching result in the vicinity of the Q, and the method of concentrating the matching results is shown below. When the arbitrary single-band image X is the matching result image 147, if the natural number g, the real number 11 and 〃 are used, the matching image 172 is updated according to Expressions 38 and 39. The matching result image 147 is a single-band image. For convenience, an image with a frequency band of 1 is used for processing. Therefore, the third added word of the function Ψ (X) is 1. φΰ · 1 (χ) 9 L eq (g, ~ 〇 + Σ 吣, m, i) iV (?) is called (5, Ami) = (eq (/ i, + ΣZρ (/, Γη, ΐ) € Ρ {; · ι ((7) and 2 {eq (g) xijl) + Ep (/, m, i) GPl; i ((?) E (l (5 ^ ^ ^ i; i for 1 < g < n and only one of g, 0 otherwise. eq (乜,
丨 1·---.------•裝--------訂--------- (請先閱讀背面之注意事項再填寫本頁) (38) (39) 在匹配結果圖像147不再變化爲止，資料處理裝置 110利用反覆計算式38及39，即能集中整個匹配結果圖像 98 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公髮） 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明（今7) 147的匹配結果。此時，藉組合規格化圖像145及模板圖 像146，匹配結果集中如下。若規格化圖像145的約一半 像素最類似特定的模板圖像146的圖像的話，則匹配結果 圖像147的整個匹配結果，集中在特定的模板圖像146的 號碼。然而，若規格化圖像145的若干像素塊類似一些不 同模板圖像146的像素塊的話，則在匹配結果圖像147中 ，能形成用0圍住的若干模板圖像146的號碼塊。再者， 若規格化圖像145與模板圖像146的集合不相關的話，則 匹配結果圖像147的匹配結果大致爲〇。因此，利用資料 處理裝置110來實現的圖形匹配，雖不易特定最類似規格 化圖像145的模板圖像146，但能從模板圖像146中選擇若 干類似的模板圖像146。因此，根據圖形匹配的結果生成 辨識結果的過程中，只要使用圖形匹配所生成的匹配結果 圖像147所列舉的模板圖像146的類似候補中，選擇一個 最有力的候補即可。 如第27圖所示，利用資料處理裝置11〇所實現的申請 專利範圍第11項之圖形匹配，爲了生成模板圖像146中從 顯示與規格化圖像145最類似的圖像號碼的匹配結果所構 成的匹配結果圖像147，網格狀排列的陣列運算單元40是 同步並列進行動作。設格子上1行」列所配置的陣列運算 單元40爲AOUu，ΑΟΙΙϋ的算法是如第28圖所示。 在步驟2901中，將A〇Uu配置在格子上的1行j列中 。無論在邏輯上或物理上，此係爲了決定A0UU的附近所 需者。 99 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐) (請先閱讀背面之注意事項再填寫本頁) 裝---- 訂--- 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明（货） 〆 在步驟2902中，設定AOUu的附近與變數的初始値。 在附近設定時，可各別決定前述各函數所使用的附近尺寸 q，亦可全部加以統一。本發明的資料處理裝置110，爲了 提高所生成的匹配結果圖像147的正確性，最好是能將附 近大小q全部設定爲較大的値。但若因此更新匹配結果所 使用之計算時間的限制、與輸入的規格化圖像145的尺寸 等因素，圖形匹配亦可視需要適當變更附近尺寸，以進行 處置。 在步驟2903中，判斷依序輸入的模板圖像146是否已 經沒有。若無模板圖像146的話(步驟2903 : YES)，即移到 步驟2905。若有模板圖像146的話(步驟2903 : NO)，即移 到步驟2904。 在步驟2904中，輸入模板圖像146的1行j列像素頻 帶數部分。因此，AOUu的至少需要記憶乘上頻帶數與模板 圖像146的數部分的圖像資料的記憶體42。之後，回到步 驟.2903。 在步驟2905中，判斷依序輸入的規格化圖像145是否 已經沒有。若無有規格化圖像145的話(步驟2905 ·· YES)， 即結束算法。若有規格化圖像145的話(步驟2905 : N〇）， 即移到步驟2906。但，當僅對特定的圖像尺寸組裝陣列運 算單元40時，作爲無限回路亦可。 在步驟2906中，輸入規格化圖像145的1行」列像素 頻帶數部分。因此，AOUu的至少需要記憶頻帶數部分的圖 像資料的記憶體42。 100 本紙張尺度適用中國國家標準（CNS)A4規格C210 X 297公釐） --»---Γ I I I I I --------訂--------- (請先閱讀背面之注意事項再填寫本頁) 經濟部智慧財產局員工消費合作社印制衣 540010 Α7 Β7 五、發明說明（ι1) 〆 在步驟2907中，從規格化圖像145與模板圖像146， 計算匹配結果圖像147的匹配結果5 υι。匹配結果爲顯示最 近規格化圖像145的模板圖像146的號碼的頻帶像素値。 在步驟2908中，AOUu是利用與附近的陣列運算單元 4〇之通信，對匹配結果圖像147的各頻帶像素値，根據函 數Ψ4χ)更新匹配結果。所更新的頻帶像素値，再被作爲 匹配結果圖像的頻帶像素値加以處理。此處，反覆函數Ψ 4x)在本來匹配結果圖像147的頻帶像素値不再變化爲止 ，反覆進行。但是，若因爲計算時間的限制、所輸入之規 格化圖像145的品質、更新之匹配結果圖像147所求的品 質等，更新處理最好是能以適當的反覆次數結束算法。 在步驟2909中，輸出匹配結果圖像147的頻帶像素値 。之後，回到步驟2905。 因此，使用由陣列運算單元40所構成資料處理裝置 110 ’對應圖像辨識機構29中的圖形匹配的申請專利範圍 第11項之視覺裝置，能從規格化圖像145生成匹配結果圖 像 147。 &上’已就使用由陣列運算單元40所構成資料處理裝 置11〇’進行僅由附近處理所構成的圖像處理的方法加以 說明°以下，使用由陣列運算單元40所構成資料處理裝置 110 ’就僅用附近處理，實現物體/背景分離機構16(參照第 2圖及第7圖）的申請專利範圍第11項之視覺裝置加以說明 〇 首先1 ’非線性振邊器一般會引起拉入現象。該拉入現 101 --^---U--------------^--------- (請先閱讀背面之注意事項再填寫本頁) 本紙張尺度適用中國國家標準（CNS)A4規格（21〇 χ 297公髮） 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明（/π) 象，是在限制循環(lmnt cycle)與吸引器(attractei·)等的周期 動作中，具有不同周期的非線性振盪器相互動作以簡單的 常數比周期進行振動，而加以限制的現象。此時，若使一 個非線性振盪器變化的話，由於其他非線性振盪器的振動 也配合變化，故該等非線性振盪器係同步。而且，藉調整 非線性振盪器的相互作用，能儘量減小或加大彼此振動的 相位差。因此，若操作此相互作用，即能將非線性振盪器 的一群分割成具有不同相位的許多組。物體/背景分離機構 16即係利用此非線性振盪器的拉入現象，將邊緣資訊圖像 中的邊緣資訊作爲父界，分離物體與背景，以生成顯示物 體領域的物體領域圖像。又，此處，舉使用范德桿(Vander Pole)作爲非線性振盪器之情形加以說明。 首先’在由網格狀排列的非線性振盪器所構成的非線 性振盪器網路中’假設位於i行j列的非線性振盪器爲从」 •，位於非線性振盪器爲〇^的Q附近的非線性振盪器的集合 Ωυ(ς) ’即目匕以式40加以顯不。但，q是4、8、24、48、 80、120、（2r + 1)2- 1的連續序列，r是自然數。又，當超 出網路尺寸的非線性振盪器包含在附近集，即以 非線性振盪器ωυ加以取代。據此，邊緣處理即能自動進行 。因此，附近集合Q“(q)之要素的數量經常爲q。又，由此 可知’非線性振盪器網路係與單頻帶圖像相同之方式加以 處理。爲了簡單地表現，在非線性振盪器網路中所添加字 僅使用寬方向與高方向兩者。 102 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公爱） -----^--------------^--------- (請先閱讀背面之注意事項再填寫本頁) 540010丨 1 · ---.------ • Install -------- Order --------- (Please read the precautions on the back before filling this page) (38) ( 39) Until the matching result image 147 no longer changes, the data processing device 110 can use iterative calculation formulas 38 and 39 to concentrate the entire matching result image 98 This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 Published by the Ministry of Economic Affairs, Intellectual Property Bureau, Consumer Cooperatives, printed 540010 A7 B7 V. Explanation of the invention (now 7) 147 matching result. At this time, by combining the normalized image 145 and the template image 146, the matching result set is as follows. If about half the pixels of the normalized image 145 are most similar to the image of the specific template image 146, the entire matching result of the matching result image 147 is concentrated on the number of the specific template image 146. However, if the pixel blocks of the normalized image 145 are similar to the pixel blocks of different template images 146, in the matching result image 147, number blocks of several template images 146 surrounded by 0 can be formed. Furthermore, if the normalized image 145 and the set of template images 146 are not related, the matching result of the matching result image 147 is approximately zero. Therefore, it is not easy to specify the template image 146 that is most similar to the normalized image 145 for the pattern matching using the data processing device 110, but it is possible to select several similar template images 146 from the template image 146. Therefore, in the process of generating a recognition result based on the result of the pattern matching, as long as the similar candidates of the template image 146 listed in the matching result image 147 generated by the pattern matching are used, one of the most powerful candidates may be selected. As shown in FIG. 27, the pattern matching of item 11 of the patent application range realized by the data processing device 11 is used to generate a matching result from the image number of the template image 146 that is most similar to the normalized image 145. The resulting matching result image 147 is operated in parallel by the array operation unit 40 arranged in a grid pattern. Suppose that the array operation unit 40 arranged in a “row” on the grid is AOUu, and the algorithm of AIOI is shown in FIG. 28. In step 2901, AoUu is arranged in row j of column 1 on the grid. Whether logically or physically, this is to determine what is needed near A0UU. 99 This paper size is in accordance with Chinese National Standard (CNS) A4 (210 X 297 mm) (Please read the precautions on the back before filling this page) Print 540010 A7 B7 V. Description of the invention (goods) 〆 In step 2902, set the vicinity of AOUu and the initial value of the variable. In the neighborhood setting, the nearby dimensions q used by the aforementioned functions may be individually determined, or all of them may be unified. In the data processing device 110 of the present invention, in order to improve the accuracy of the generated matching result image 147, it is preferable to set all the nearby sizes q to a large value. However, if the calculation time used for the matching result is limited due to such factors as the size of the input normalized image 145 and other factors, the graphic matching may also be appropriately changed in the nearby dimensions for disposal. In step 2903, it is determined whether or not the template images 146 which have been sequentially inputted are not present. If there is no template image 146 (step 2903: YES), the process moves to step 2905. If there is a template image 146 (step 2903: NO), it moves to step 2904. In step 2904, the pixel band number of one row and j columns of the template image 146 is input. Therefore, AOUu needs at least a memory 42 that stores image data multiplied by the number of frequency bands and the number of portions of the template image 146. After that, go back to step 2903. In step 2905, it is determined whether or not the normalized image 145 which has been sequentially inputted is no longer present. If there is no normalized image 145 (step 2905 · YES), the algorithm ends. If there is a normalized image 145 (step 2905: No), the process moves to step 2906. However, when the array operation unit 40 is assembled only for a specific image size, it may be used as an infinite circuit. In step 2906, the number of pixel bands in one row and one column of the normalized image 145 is input. Therefore, AOUu needs at least a memory 42 for storing image data in the frequency band portion. 100 This paper size applies to China National Standard (CNS) A4 specification C210 X 297 mm)-»--- Γ IIIII -------- Order --------- (Please read the back first Please pay attention to this page and fill in this page again) Printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 510010 Α7 Β7 V. Description of the invention (ι1) 〆 In step 2907, calculate the matching result from the normalized image 145 and the template image 146 Image 147 matches 5 υι. The matching result is a band pixel 値 displaying the number of the template image 146 of the most recent normalized image 145. In step 2908, AOUu uses the communication with the nearby array operation unit 40 to update the matching result for each band pixel 値 of the matching result image 147 according to the function Ψ4χ). The updated band pixels 値 are processed as band pixels 图像 of the matching result image. Here, the iteration function (4x) is repeated until the frequency band pixels 匹配 of the original matching result image 147 no longer change. However, if the calculation time is limited, the quality of the input normalized image 145, the quality of the updated matching result image 147, etc., the update process should preferably end the algorithm with an appropriate number of iterations. In step 2909, the band pixels 値 of the matching result image 147 are output. After that, it returns to step 2905. Therefore, by using the visual processing device 110 of the patent application scope item 11 corresponding to the pattern matching in the image recognition mechanism 29, the data processing device 110 'composed of the array operation unit 40 can generate a matching result image 147 from the normalized image 145. & The method of performing image processing consisting of only nearby processing using the data processing device 11 composed of the array operation unit 40 has been described above. Hereinafter, the data processing device 110 composed of the array operation unit 40 is used. 'Just use the nearby processing to realize the visual device of the 11th patent application scope of the object / background separation mechanism 16 (refer to Figure 2 and Figure 7). First of all,' Nonlinear edger generally causes pull-in. phenomenon. The pull-in is now 101-^ --- U -------------- ^ --------- (Please read the notes on the back before filling this page) Paper size applies Chinese National Standard (CNS) A4 specification (21〇χ 297 issued) Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 540010 A7 B7 V. Description of the invention (/ π) The phenomenon is in the lmnt cycle In a periodic operation such as an attractor, a phenomenon in which a non-linear oscillator having a different cycle interacts with a simple constant specific period to vibrate and is restricted. At this time, if one non-linear oscillator is changed, since the vibration of other non-linear oscillators also changes in coordination, the non-linear oscillators are synchronized. Moreover, by adjusting the interaction of the non-linear oscillators, it is possible to minimize or increase the phase difference between the vibrations of each other. Therefore, if this interaction is operated, a group of non-linear oscillators can be divided into many groups with different phases. The object / background separation mechanism 16 uses the pull-in phenomenon of the non-linear oscillator, and uses the edge information in the edge information image as the parent boundary to separate the object and the background to generate an image of the object area showing the object area. Here, a case where a Vander Pole is used as a nonlinear oscillator will be described. First, 'in a nonlinear oscillator network composed of grid-shaped nonlinear oscillators', it is assumed that the nonlinear oscillators located in row i and column j are slaves ", and Q located in the nonlinear oscillator is 0 ^ The set of nearby non-linear oscillators, Ωυ (ς) ', is shown in Equation 40. However, q is a continuous sequence of 4, 8, 24, 48, 80, 120, (2r + 1) 2-1, and r is a natural number. In addition, when the non-linear oscillator exceeding the network size is included in the nearby set, it is replaced by the non-linear oscillator ωυ. Based on this, edge processing can be performed automatically. Therefore, the number of elements in the nearby set Q "(q) is often q. From this, it can be seen that the 'non-linear oscillator network is processed in the same way as a single-band image. For simple expression, the nonlinear oscillation The words added in the computer network only use both the wide and high directions. 102 This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 public love) ----- ^ ------- ------- ^ --------- (Please read the precautions on the back before filling out this page) 540010
五、發明說明（/c ί^ζ + Uhj- (¢) if 9 = 4, (40) \i-r <1 <ι + τ, j - r <m<j + r, ω1ιΤΪΙ φ ω^} if g = (2r + l)2 — 1* 其次’非線性振盪器，係與位於qa附近所含的附近集 合Ωϋ(ς〇的非線性振盪器之間，利用式41所計算的結合値 ruu進行結合。又，當不使用對數表時，亦可使用式42白勺 近似。另外，#、y是適當正的常數。 ^ijki = μ sinc((z^fe)2t(j ~—) for Va,-W G Ω^(ζ?α) u2 (41) sinc(x) ^ < 1 一 2|x|2 + M3 if〇S|x|<l， 4 一 8|x| + 5|o;|2 — |x|3 迁1$ |x| <2, 〇 otherwise. (42) 經濟部智慧財產局員工消費合作社印製 當非線性振盪器網路之所有非線性振盪器係完全同相 位且同步時，只要是以處理器進行計算，非線性振盪器ωυ 即會在永遠在同相位的狀態下持續動作。因此，若能賦予 千擾Ρυ的話就能回避這種狀態。作爲此干擾，雖能使用擬 似隨機數，但使用如式43般簡單的式加以求出即足夠。又 5 Γυ是顯示邊緣資訊圖像的i行」列的邊緣資訊的有無。 若有邊緣資訊的話，則爲1，若無邊緣資訊的話，則爲〇。 此外’ k是適當之正的常數。V. Description of the invention (/ c ί ^ ζ + Uhj- (¢) if 9 = 4, (40) \ ir < 1 < ι + τ, j-r < m < j + r, ω1ιΤΪΙ φ ω ^ } if g = (2r + l) 2 — 1 * Secondly, the non-linear oscillator is a combination of the non-linear oscillator located near qa with the nearby set Ωϋ (ς0), calculated using Equation 41. ruu is combined. When a logarithmic table is not used, an approximation of Equation 42 can also be used. In addition, # and y are appropriate positive constants. ^ ijki = μ sinc ((z ^ fe) 2t (j ~ —) for Va, -WG Ω ^ (ζ? α) u2 (41) sinc (x) ^ < 1-2 | x | 2 + M3 if〇S | x | < l, 4-8 | x | + 5 | o; | 2 — | x | 3 Move 1 $ | x | < 2, 〇otherwise. (42) All non-linear oscillator systems of the non-linear oscillator network printed by the Consumer Cooperative of Intellectual Property Bureau of the Ministry of Economic Affairs When completely in phase and synchronized, as long as the calculation is performed by the processor, the non-linear oscillator ωυ will continue to operate in the state of the same phase forever. Therefore, if you can give perturbation to Pυ, this state can be avoided. As Although this interference can use quasi-random numbers, it can be added using a simple equation like Equation 43. It is enough to find it. And 5 Γυ is the presence or absence of edge information in the "row i" column of the edge information image. If there is edge information, it is 1, and if there is no edge information, it is 0. In addition, 'k is appropriate Positive constant.
Pij ^ (43) 爲了使非線性振盪器爲ωυ與附近集合Qu(qa)的非線性 103 本紙張尺度適用中國國家標準（CNS)A4規格《210 X 297公釐） -------------------訂--------- (請先閱讀背面之注意事項再填寫本頁) 540010 Α7 Β7 五、發明說明（z 二 Σ + (〜·): ^kieflij(qb) (48) 以此結果爲基準，利用式49計算顯示邊緣資訊的內插 比例的交界參數Γυ。又，α、/3、7ymin、；7max是適當之正 的常數。 請 先 閱 讀 背 dt if Vmin ^ Vij — Vmaxt (49) /3(1 — ξ^) otherwise. 經濟部智慧財產局員工消費合作社印製 以上，雖已就作爲非線性振盪器之范德桿的情形加以 說明，但除此之外以布魯塞爾機(Brusselator)般之限制循環 且穩定的非線性振盪器，或產生洛倫茲吸引器與雷斯拉方 程式的吸引器的無秩序振盪器等，能引起拉入現象之任意 的非線性形振盪器也能動作。在此情況下，只要將參數 與Ψυ以非線性形振盪器加以置換追加即可。此時，僅係在 適當的參數中加入附近輸入合計σ u與干擾ρ &。但，若係 無秩序振盪器時，並不需要干擾。 使用式40到式49，可記述能組裝物體/背景分離機構 16(黎照第2圖及桌7圖）之資料處理裝置iiq的所有陣列運 算單元40的算法。以下，使用資料處理裝置u〇中的任意 陣列運算單元40的算法，就對應物體/背景分離機構丨6的 申請專利範圍第12項之視覺裝置，加以說明。 如桌29 Η所不，使用以資料處理裝置HQ所實現的物 體/背景分離機構16所形成的三角形的邊緣資訊，使分 離成三角形的內側領域152與三角形的外側領域153，網 105 本纸張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 意 事 項 再 填 寫 本 頁 訂 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明 格狀排列的陣列運算單元40是同步並列動作。設格子上χ 行j列所配置的陣列運算單元40爲AOUu，則AOUu的算法 如第30圖所示。 在步驟1601中，將AOUu配置在格子上的！行列。 在步驟1602中，根據式41及42，用結合値連接 ω ϋ 與 ω η 〇 在步驟1603中，於非線性振盪器的參數如與1|^設定 適當的値。 在步驟1604中，判斷依序輸入的形成邊緣資訊圖像 115是否已經沒有。若無形成邊緣資訊圖像丨15的話(步驟 1604 : YES) ’即結束算法。若有形成邊緣資訊圖像ι15的 話(步黩1604 : NO)，即移到步驟16〇5。但，當僅對特定的 頻帶數及圖像尺寸組裝陣列運算單元40時，作爲無限回路 亦可。 在步驟1605中，輸入形成邊緣資訊U4的。 在步驟1606中’從前一刻輸入的形成邊緣資訊U4的 Γυ，根據式43計算干擾Ρϋ。 在步驟1607中，自具有附近集合Qu(qa)中的非線性振 盪器之陣列運算單元40之AOUu，輸入、fu、Wkl ，根據式44計算合計値σ u。 在步驟1608中，根據式45及46，計算非線性振盪器 的參數Φυ與Ψυ。亦即，以尤格庫塔(Runge Kutta)法，解該 等式所顯示的微分方程式。 在步驟1609中，根據式47，計算非線性振盪器的輸 106 ^^尺度適用中國國家標準（CNS)A4規格（210 X 297公爱 1 ---------------裝--------訂------ (請先閱讀背面之注意事項再填寫本頁) 線·！ 540010 經濟部智慧財產局員工消費合作社印製 Α7 Β7 五、發明說明（/$) ^ 出λ。。此處，若ΨΘ 0的話，則λϋ=卜除此之外，；U = 〇。 在步驟16010中，自具有附近集合Qu(cib)中之非線性 振盪器〇kl的陣列運算單元40的AOUu輸入Au，根據式 48計算輪廓參數u。 在步驟1611中，根據式49計算交界參數f η。亦即， 用差分法或尤格庫塔(Runge Kutta)法，解該等式所顯示的 微分方程式。 在步驟1612中，判斷顯示步驟1606到步驟1611的反 覆次數的分離次數，是否已達到指定次數。若分離次數沒 有達到指定次數的話(步驟1612 : NO)，即回到步驟1606。 若分離次數達到指定次數的話(步驟1612 : YES)，即移到步 驟 1613 。 在步驟1613中，輸出作爲物體領域圖像142之頻帶像 素値的非線性振盪器的輸出λυ。之後，回到步驟1604。 又，爲了求步驟1612中之分離次數’能使用如下的方 法。在物體/背景分離機構16中，若圖像尺寸是一定的話 ，由於與非線性振盪器的初始狀態無關的，在大致所有的 形成邊緣資訊Η4中經一定時間即結束分離，故預先計算 此時間求出步驟到步驟1611之反覆次數即可。此係 因若非線性振盪器的初始狀態在一定範圍內的話，因拉入 現象的影響’非線性振盪器達到同步的時間沒有很大差異 的關係。 因此，僅計算非線性振盪器，即能使用形成之三角形 107 1纸張尺度 + _家標準（CNS)A4規格⑽Χ 297公^ " --»---r------裝--------訂---------線^1- (請先閱讀背面之注意事項再填寫本頁) 540010 A7 B7 經濟部智慧財產局員工消費合作社印製 五、發明說明（/^) 的邊緣資訊151分離三角形的內側領域152與三角形的外 側領域153，是因爲利用非線性振盪器之性質的拉入現象 的關係。亦即，當用正的結合値來結合兩個非線性振盪器 時係成爲同相位，而用負的結合値結合時則係極力成爲大 的相位差。使用此性質的話，由於係用附近彼此正的結合 値來結合網格狀並列的非線性振還器，因此不直接結合的 非線性振盪器彼此爲同相位。而且，若用負的結合値來結 合位於隔離形成邊緣資訊114位置的非線性振盪器的話， 則邊緣資訊的兩側極力錯開。以此方式進行，即使不結合 所有的非線性振盪器，在三角形的邊緣資訊151的內側與 外側亦能成爲各自不同的相位集合。因此，物體/背景分离隹 機構16，分離成第29圖所示之三角形的內側領域152與三 角形的外側領域153。此時，三角形的內側領域152與三 角形的外側領域153的相位差超過90度而盡可能的接近 180度，分離出三角形與背景領域。 此處，最重要的是，本實施形態，是用能得到形成邊 緣資訊114所不的方法，對結合値進行擬似變更。首先， 如式41及42所規疋般’將非線性振邊器〇 k 1用來結合爲非 線性振盪器ωϋ的結合値設爲 Γ ijkl (參照步驟1602)。形成邊 緣資訊Γ u與Γ kl，在具有邊緣時，皆爲1，沒有邊緣時， 皆爲0。輸入形成邊緣資訊Γ u與Γ u後(參照步驟1605)， 從陣列運算單元40的AOUu，將形成邊緣資訊rki傳送到 AOUr在AOUu計算^A(l— D以取代結合値r叫（參 照步驟1607)。對此代用的接合値rUklA(丨—f k〇，交界參 108 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） '--- (請先閱讀背面之注意事項再填寫本頁) 一裝Pij ^ (43) In order to make the non-linear oscillator ωυ and the non-linearity of the nearby set Qu (qa) 103 This paper size applies the Chinese National Standard (CNS) A4 specification "210 X 297 mm" ------- ------------ Order --------- (Please read the notes on the back before filling this page) 540010 Α7 Β7 V. Description of the invention (z 二 Σ + (~ · ): ^ kieflij (qb) (48) Based on this result, use Equation 49 to calculate the boundary parameter Γυ of the interpolation ratio showing the edge information. In addition, α, / 3, 7ymin, and 7max are appropriate positive constants. Please read back dt if Vmin ^ Vij — Vmaxt (49) / 3 (1 — ξ ^) otherwise. Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs, although it has been used as a van der pole for nonlinear oscillators Explained, but in addition to the Brussels machine (Brusselator) limited cycle and stable non-linear oscillator, or the Lorentz attractor and the Lasla equation attractor, disordered oscillator, etc., can cause pull Arbitrary non-linear oscillators can also operate. In this case, as long as the parameter and Ψυ oscillate in a non-linear shape It can be replaced and added. At this time, only the nearby input total σ u and interference ρ & are added to the appropriate parameters. However, if it is a disorderly oscillator, interference is not required. Using Equation 40 to Equation 49, you can Describe the algorithms of all the array operation units 40 of the data processing device iiq that can assemble the object / background separation mechanism 16 (Figure 2 and Table 7). Hereinafter, any array operation unit 40 in the data processing device u0 is used. The algorithm explains the visual device corresponding to item 12 of the patent application scope of the object / background separation mechanism 丨 6. As shown in Table 29, using the object / background separation mechanism 16 implemented by the data processing device HQ The edge information of the triangle separates the inside area 152 of the triangle and the outside area 153 of the triangle NET 105. This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm). Printed by the Consumer Cooperatives of the Ministry of Intellectual Property Bureau 540010 A7 B7 V. Description of the Invention The grid-shaped array operation unit 40 is synchronized and operated side by side. The array operation unit 40 arranged in column j is AOUu, and the algorithm of AOUu is as shown in Fig. 30. In step 1601, AOUu is arranged on the grid! Rows and columns. In step 1602, according to equations 41 and 42, use Combine 値 to connect ω ϋ and ω η 〇 In step 1603, set the appropriate 于 for the parameters of the non-linear oscillator such as and 1 | ^. In step 1604, it is determined whether or not the formed edge information image 115 that has been sequentially inputted is no longer present. If no edge information image is formed (step 1604: YES), the algorithm ends. If an edge information image ι15 is formed (step 1604: NO), the process moves to step 1605. However, when the array operation unit 40 is assembled only for a specific number of frequency bands and image sizes, it may be used as an infinite loop. In step 1605, input the edge information U4. In step 1606 ', Γυ forming the edge information U4 input from the previous moment, calculates the interference Pϋ according to Equation 43. In step 1607, from the AOUu of the array operation unit 40 having the nonlinear oscillator in the nearby set Qu (qa), input, fu, Wkl, and calculate a total 値 σu according to Equation 44. In step 1608, the parameters Φυ and Ψυ of the non-linear oscillator are calculated according to Equations 45 and 46. That is, the differential equation shown by this equation is solved by the Runge Kutta method. In step 1609, according to Equation 47, calculate the output 106 ^^ scale of the non-linear oscillator to apply the Chinese National Standard (CNS) A4 specification (210 X 297 public love 1 --------------). -Install -------- Order ------ (Please read the precautions on the back before filling this page) LINE ·! 540010 Printed by the Intellectual Property Bureau Staff Consumer Cooperatives of the Ministry of Economic Affairs Α7 Β7 V. Description of the invention (/ $) ^ Out of λ. Here, if ΨΘ 0, then λ 卜 = Bu besides; U = 〇. In step 16010, it has a non-linear oscillator in the nearby set Qu (cib). The AOUu of the array operation unit 40 of 〇kl inputs Au, and calculates the contour parameter u according to Equation 48. In step 1611, the boundary parameter f η is calculated according to Equation 49. That is, the difference method or the Runge Kutta method is used. , And solve the differential equation shown in the equation. In step 1612, it is judged whether the number of separations showing the repeated number of steps 1606 to 1611 has reached the specified number. If the number of separations has not reached the specified number (step 1612: NO) That is, return to step 1606. If the number of separations reaches the specified number (step 1612: YES), move to step 161 3. In step 1613, the output of the non-linear oscillator λυ which is the pixel 値 of the band of the object area image 142 is output. After that, return to step 1604. In order to obtain the number of separations in step 1612, the following method can be used In the object / background separation mechanism 16, if the image size is constant, since it has nothing to do with the initial state of the non-linear oscillator, the separation will be completed after a certain time in almost all of the formation edge information Η4, so this is calculated in advance. Time to find the number of iterations from step to step 1611. This is because if the initial state of the nonlinear oscillator is within a certain range, due to the effect of the pull-in phenomenon, there is no significant difference in the time when the nonlinear oscillator reaches synchronization. Therefore, only the non-linear oscillator is calculated, that is, the triangle 107 1 paper size + _ house standard (CNS) A4 size ⑽ 297 public ^ "-»--- r ------ installed -------- Order --------- line ^ 1- (Please read the precautions on the back before filling this page) 540010 A7 B7 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs Invention description (/ ^) of the edge information 1 51 separates the inner region 152 of the triangle from the outer region 153 of the triangle because the relationship of the pull-in phenomenon of the nature of the non-linear oscillator is used. That is, when two non-linear oscillators are combined with a positive coupling 値, it becomes In the same phase, when using a negative coupling 値, the system will try to become a large phase difference. Using this property, because the system uses a positive coupling 値 near each other to combine the grid-shaped parallel non-linear vibrators, it is not direct. The combined nonlinear oscillators are in phase with each other. Furthermore, if a non-linear oscillator located at the position where the edge information 114 is isolated is combined with a negative coupling 値, the two sides of the edge information are strongly staggered. In this way, even if all non-linear oscillators are not combined, the inside and outside of the edge information 151 of the triangle can become different sets of phases. Therefore, the object / background separation / mechanism 16 is separated into a triangular inner region 152 and a triangular outer region 153 as shown in FIG. At this time, the phase difference between the inside area 152 of the triangle and the outside area 153 of the triangle is more than 90 degrees and as close as possible to 180 degrees, and the triangle and the background area are separated. Here, the most important thing is that this embodiment uses a method that can obtain the formation edge information 114, and similarly changes the coupling. First, as shown in Equations 41 and 42, the combination 値 of the non-linear vibrator 0 k 1 to be used as the non-linear oscillator ωϋ is set to Γ ijkl (see step 1602). The edge information Γ u and Γ kl are formed, and are both 1 when there are edges, and 0 when there are no edges. After inputting the formation edge information Γ u and Γ u (refer to step 1605), from the AOUu of the array operation unit 40, the formation edge information rki is transmitted to the AOUr and the AOUu is calculated ^ A (l- D instead of the combination 値 r (refer to step 1607). For this alternative splicing 値 rUklA (丨 —fk〇, junction reference 108 This paper size applies to China National Standard (CNS) A4 specifications (210 X 297 mm) '--- (Please read the precautions on the back first (Fill in this page again)
· I I I I •線•丨 經濟部智慧財產局員工消費合作社印製 540010 Α7 Β7 五、發明說明（/^) 數Γυ是作爲從〇到1的倍率作用（參照步驟1607)。 如第31圖所示，當形成邊緣資訊114爲虛線狀態的三 角形的邊緣資訊154時，必須進行虛線的內插。首先，當 使用虛線狀態的三角形的邊緣資訊154使系統動作時(參照 步驟1605)，在虛線狀態的三角形的邊緣資訊154的內側與 外側相位差雖約超過90度，但三角形的內側與外側的交界 部分則不明確，因此，各AOUu計算非線性振盪器的輸出 λ u(參照步驟1609)。當此輸出λ ϋ爲1時，在附近的非線性 振盪器中設Au爲1時的非線性振盪器爲，則參數Ψϋ與 Ψ kl问時爲Θ以上。亦即’ λ ϋ與λ kl約爲同相位，若Θ爲正 値的話，最差時相位差也不至超過90度。該相位差的最大 値是藉Θ値來決定，在λ ϋ與；I u同時爲1的範圍下若加大 0的話，該相位差即趨近0度。因此，當使用λϋ與Au的 話，在附近的非線性振盪器中，顯示約略同相位數的輪廓 參數7/ u是根據式48進行計算(參照步驟1610)。接著，若 該輪廓參數7? ϋ在整個附近中，約爲一半的話，則根據式 49減少結合値的倍率的交界參數，除此以外，則根據式 49使之增加(參照步驟1611)。例如，在8附近時，若是3 到5之間，根據式49減少交界參數即即可。若反覆該過程 使之持續動作的話，當得到第31圖所示的虛線狀態的三角 形的邊緣資訊154時’即分離爲虛線三角形的內側領域 155與虛線三角形的外側領域156。 如第32圖所示，當重疊兩個三角形時，能得到前方三 角形的邊緣資訊157與後方三角形的邊緣資訊158。此時 109 本紙張尺度適用中國國家標準（CNS)A4規格（210 χ 297公爱） (請先閱讀背面之注意事項再填寫本頁) -裝 ——訂---------線· 經濟部智慧財產局員工消費合作社印製 540010 Α7 Β7 五、發明說明（/θ) · ，利用前方三角形的內側領域159與後方三角形的內側領 域160與雙重三角形的背景領域161的3個領域的非線性 形振盪器之相位的相互偏離，分離成三個領域。又，如第 33圖所示，兩個重疊圓形的邊緣資訊162即使是虛線，也 被分離爲前方圓形的內側領域163與後方圓形的內側領域 164與雙重圓的背景領域165的三個領域。 因此，使用由陣列運算單元40所構成的資料處理裝置 110，以對應物體/背景分離機構16的申請專利範圍第12 項之視覺裝置，能以形成邊緣資訊圖像115的形成邊緣資 訊114爲交界，分離物體領域141與背景。 以上，就申請專利範圍第6項到申請專利範圍第12項 之視覺裝置，進行了說明。當然，該等視覺裝置雖能利用 一般電腦進行組裝，但當將移動物體2作爲計數對象時， 必須根據移動物體2的移動速度高速地執行各個前述機構 。尤其是當提高框圖像1的圖像尺寸或解像度時，以圖像 本身作爲處理對象的圖像記憶機構12(參照第1圖及第6圖 )、圖像振動機構13(參照第3圖、第4圖及第5圖）、邊緣 資訊生成機構14(參照第1圖及第6圖）、邊緣資訊形成機 構15(參照第1〜5圖、第7〜12圖）、物體/背景分離機構 16(參照第2、4、5、7〜12圖）、位置/大小檢測機構17(參 照第1圖及第6圖）、領域規格化機構27(參照第7圖）、規 格化圖像保持機構28(參照第7圖)及圖像辨識機構29(參照 第8圖），對各寬方向及各高度方向之計算量，即與圖像尺 寸或解像度成正比增大。因此，申請專利範圍第6項到申 110 本紙張尺度適用中國國家標準（CNS)A4規格(210 X 297公釐） -------------裝--------訂--------- (請先閱讀背面之注意事項再填寫本頁) 經濟部智慧財產局員工消費合作社印製 540010 A7 ____ B7 五、發明說明以卞 請專利範圍第12項之視覺裝置，有時可能會因用途之不同 ，而無法達到預期的性能。 此處，爲了使用數位技術組裝圖像記憶機構12、圖像 振動機構13、邊緣資訊生成機構14、邊緣資訊形成機構 15、物體/背景分離機構16、位置/大小檢測機構17、領域 規格化機構27、及規格化圖像保持機構28，申請專利範圍 第13項之陣列運算單元40是在資料處理裝置110中，如 第13圖所示般排列爲網格狀，而且，陣列運算單元40係 以僅能與資料處理裝置110中的相鄰陣列運算單元40相互 通信之方式，加以配線。亦即，4附近彼此間係直接配線 。因此，與8彼此間配線時比較起來，使用較少的電子零 件與配線量，就能同程度高速地動作，且能在將來欲擴充 附近尺寸時具有簡單地擴充性。 申請專利範圍第14項之陣列運算單元40，入第34圖 所示，係由用來計算圖樣處理之式的處理器(Processor)41、 用來記憶式中所使用的所有參數、常數、函數及運算子的 記憶體(Mem〇ry)42、以及用來與附近的陣列運算單元40進 行通信的控制器(Contr〇ler)43所構成，處理器41可利用位 址匯流排51所指定的位址(Address)，選擇記憶體42及控 制器43的任意記憶元件及暫存器。此外，處理器41是透 過位址匯流排51連接於記憶體42及控制器43以能雙向通 信，以存取位址匯流排51所指定的任意記憶體元件及暫存 器的資料(DATA)。當陣列運算單元40輸入由一個以上的 輸入像素所構成的前輸入資料群(Front Input Data Set)時， 111 本紙張尺度適用中國國家標準（CNS)A4規格（210 χ 297公釐） I -----------裝--------訂---------^9— (請先閱讀背面之注意事項再填寫本頁) 540010 A7 B7 五、發明說明（丨0) 控制器43即將前輸入資料群記憶在記憶體42。又，控制 器43將由函數所作成的記憶體42中的計算資料傳送到相 鄰的陣列運算單元40,同時將從相鄰的陣列運算單元40 所接收的計算資料記憶在記憶體42，若有需要，傳送到輸 入以外的陣列運算單元40。最後，控制器43將輸出圖像 資料作爲結果資料(Result Data)加以輸出。 如上所述，在陣列運算單元40中搭載控制器43的理 由，是由於在陣列運算單元40彼此通信期間處理器41亦 能動作，故處理器41在通信的等待時間中也能進行計算而 實現高速處理，及即使變化附近的陣列運算單元40的數量 也不必變更硬體，以及控制器43進行圖像的邊緣處理，亦 即，能自動進行對圖像中之邊緣像素的例外處理，故處理 器41的程式不必進行邊緣處理而極爲單純之故。 經濟部智慧財產局員工消費合作社印製 -----*------•裝--------訂 (請先閱讀背面之注意事項再填寫本頁) 處理器41與記憶體42能使用一般的數位電路。控制 器43.的具體電路圖是如第35圖所示。位址緩衝器(Address Buffer〇53是透過位址匯流排(Address Bus)5：l，從處理器41 接收位址(Address) ’根據位址解碼器(Address Decoder)54， 選擇各暫存器及其他機能方塊。資料緩衝器(Data Buffer)55 ，經由資料匯流排(Data Bus)52，從處理器41接收資料 (Data)，透過用位址解碼器54所選擇的暫存器與內部資料 匯流排56進行排他的通信。通信方向藉讀取(Read)加以指 定。當位址指定標記暫存器(Flag Reg1Ste:r)57時，資料被記 憶在標§己暫存器57中，利用標記解碼器(Flag Decoder)58 進行解碼，以複數信號(Signals)，傳送到相鄰的陣列運算單 112 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 540010 A7 B7___ 五、發明說明（/(/) (請先閱讀背面之注意事項再填寫本頁) 元40。複數信號是利用標記編碼器(Flag Encoder〇59加以接 收，經解析後，記憶在狀態暫存器(Status Register)60中， 又，作爲接收(RECIEVER)送回傳送源的陣列運算單元40。 接收是用複數信號的傳送源的標記編碼器59來接收，其結 果，確認複數信號的傳送結束。當用位址選擇狀態暫存器 時，狀態暫存器60的內容即透過資料匯流排52以資料傳 送到處理器41。當標記編碼器59接收對應一個以上的輸 入圖像（Input Image)的一個以上的前輸入送達（Front Input Sent)時，則由一個以上輸入圖像所構成的前輸入資料群 (Front Input Data Set)，讀入準備有所需之記憶容量分的前 輸入資料暫存器(Front Input Data Register)61。當使用位址 來選擇前輸入資料暫存器61時，前輸入資料暫存器61的 內容，以資料傳送到處理器41。當處理器41完成計算後 ，根據位址選擇結果資料暫存器(Result Data Register)62， 結果資料暫存器62即將輸出圖像的圖像資料作爲結果資料 (Result Data)加以讀取。與此同時，標記編碼器59傳送結 果送達(Result Send)。 經濟部智慧財產局員工消費合作社印製 當自附近的陣列運算單元40求出計算所需的資料後， 即選擇輸出資料暫存器(Output Data Register)63作爲位址， 將應傳送到附近的陣列運算單元40的資料作爲計算資料 (Calculation Data)，讀入輸出資料暫存器63。之後，作爲計 算資料傳送至相鄰的所有陣列運算單元40。當從上側的陣 列運算單元40接收複數信號(Signals)後，即將計算資料讀 入上輸入資料暫存器(Upper Input Data Register)64。之後， 113 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明（/u) 根據位址選擇上輸入資料暫存器64後，即將上輸入杳料暫 存器64的內容作爲計算資料加以傳送。從下側、左側、右 側的陣列運算單元40接收複數信號時亦相同，下輸入資料 暫存器65、左輸入資料暫存器66、右輸入畜料暫存器67 皆同樣地動作。 各種緩衝器、各種暫存器、位址解碼器54的各方塊是 一般的電子電路。具體說來，標記解碼器58與標記編碼器 59具有第36圖與第37圖所示的輸入輸出信號。種類 (Type)疋以5位兀藏不f目ί』出資料暫存器（〇utpUt Data Register)63所讀入的內容種類。該位元數係足以區別陣列 運算單元40所收發之所有計算資料的値。計數器一 X(Count-X)及計數器一 Y(Count-Y)係顯示各個無4位元編碼 之整數，顯示陣列運算單元40間的傳送次數。當陣列運算 單元40傳送計算資料時，各個計數器爲〇，當再度傳送從 左右的陣列運算單元40所傳送的計算資料時，是在標記編 碼器59的計數器-X中加1的値，當再度傳送從上下的陣 列運算單元40所傳送的計算資料時，是在標記編碼器59 的計數器-Y中加1的値。處理器41是於標記暫存器57 的送達標記(Send Flag)中，指定輸出資料暫存器63的內容 是傳送至上下左右中那一任何方向後，當標記解碼器58接 收指定輸出資料暫存器63的位址解碼器54的中央解碼 (Central Decoding)時，標記解碼器58即將送達(Send)配合 送達標記的指定方向加以輸出。送達標記是用4位元顯示 ，當陣列運算單元40的計算資料傳送到四方的陣列運算單 114 $紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公楚） I ----------丨^^ 裝----I I--訂--------- (請先閱讀背面之注意事項再填寫本頁) 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明（//)) . 元40時，處理器41係設定爲1111 ’當將從右側的陣列運 算單元40所傳送來的計算資料傳送到上下左右側時，係設 定爲1101，當從下側傳送到上側時，設定爲，當從上 側傳送到下側時，設定爲0100。以此方式，由於不僅不至 重複傳送而能有效地傳送，而且傳送方向的決定規則變的 明確，故藉組合種類、計數器-X及計數器- Y，標記編碼 器59能判定是從哪一個陣列運算單元40、傳送哪一種的 計算資料。將計算資料作爲結果資料讀入結果資料暫存器 62中之同時，標記解碼器58亦接收結果解碼（Result Decoding)，傳送結果送達(Result Send)。 當標記編碼器59從四方之任一方向接收到送達時，即 接收接收方向的種類與計數器-X、計數器-Y，更新該部 分的狀態暫存器60的內容。在此更新的同時，將接收設定 爲1傳送至接收方向。傳送源的陣列運算單元40的標記編 碼器5·9，是在接收成爲1的瞬間進行接收，更新狀態暫存 器60的接收狀態(Receive Status)。因此，各陣列運算單元 40中，處理器41僅檢查狀態暫存器60的接收狀態，即能 判斷有效的計算資料是記憶在哪一個輸入資料暫存器中。 因此，例如，若將計算資料讀入上輸入資料暫存器64的話 ，雖然處理器41可藉指定位址從上輸入資料暫存器64讀 入資料，但同時將上解碼(Upper Decoding)從位址解碼器54 傳送到標記編碼器59，在接收狀態中之上部分回到0，將 向上側的接收作爲0加以傳送。下左右側之情形時也同樣 地動作。當標記編碼器59即使只接收到一個輸入圖像用的 115 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） --------------裝---- (請先閱讀背面之注意事項再填寫本頁) 訂-------- 540010 A7 B7 五、發明說明（/φ (請先閱讀背面之注意事項再填寫本頁) 前輸入送達，即將狀態暫存器60中對應接收前輸入送達的 輸入圖像用的前輸入送達狀態(Front Input Send Status)爲1 。又，當處理器41從輸入圖像用的前輸入資料暫存器61 讀取資料時，位址解碼器54是將前解碼(Front Decoding)傳 送到標記編碼器59，將接收的前輸入送達所對應的前輸入 送達狀態設定爲〇。處理器41藉讀取狀態暫存器60的內 容，即判斷前輸入資料暫存器61中是否記憶有最新的輸入 圖像。 經濟部智慧財產局員工消費合作社印製 處理器41透過控制器43將計算資料傳送給四方的陣 列運算單元40時的算法顯示於第38圖。第38圖是顯示使 用處理器41的程式控制，與使用標記解碼器58及標記編 碼器59的硬體邏輯的混合處理。第38圖中，步驟71，係 處理器41讀取狀態暫存器60的內容。步驟72，係判斷讀 取的內容中接收狀態是否全部爲0。若是NO的話，即結束 處理，若是YES的話，即移到步驟73。在步驟73中，處 理器41決定傳送到相鄰的陣列運算單元40的資料種類與 計數器與傳送方向，將其內容寫入標記暫存器57。在步驟 73中，處理器41是將相鄰陣列運算單元40所傳送的資料 寫入輸出資料暫存器63中。在步驟75中，將輸出資料暫 存器63的內容作爲計算資料，傳送給相鄰之陣列運算單元 40。在步驟76中，僅在標記暫存器57的送達標記所指定 的方向，將送達設定爲1加以傳送。藉此，處理器41的1 次傳送便告結束。處理器41，當每一次應傳送的資料在記 憶體42內更新時，即開始該傳送算法。 116 本紙張尺度適用中國國家標準（CNS)A4規格(210 X 297公釐） 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（/(Γ) 控制器43從上側的陣列運算單元40接收計算資料時 之算法顯示於第39圖。第39圖是顯示使用標記解碼器58 及標記編碼器59的硬體邏輯的處理。第39圖中，步驟81 ，係標記編碼器59輸入送達。在步驟82中，標記編碼器 59判斷送達是否爲1。若是NO的話，即結束處理，若是 YES的話，即移到步驟83。在步驟83中，上輸入資料暫 存器64讀取從上側所傳送來的計算資料。在步驟84中， 標記編碼器59將狀態暫存器60中上側用的接收狀態設定 爲1，同時將接收設定爲1傳送給上側的陣列運算單元40 。上下右側的情形亦相同。據此，控制器43的1次接收算 法便告結束。控制器43隨時監視來自上下左右陣列運算單 元40的送達，每當接收到該送達時即開始該接收算法。 處理器41從上輸入資料暫存器64接收資料時之算法 顯示於第40圖。第40圖是顯示使用處理器41的程式控制 ，與使用標記解碼器58及標記編碼器59的硬體邏輯的混 合處理。第40圖中，步驟91，處理器41讀取狀態暫存器 6〇的內容。在步驟92中，判斷讀入的內容中上側用的接 收狀態是否爲1。若是NO的話，即結束處理，若是YES 的話，即移到步驟93。在步驟93中，處理器41從上輸入 資料暫存器64讀取資料。在步驟94中，標記編碼器59將 狀態暫存器60中上側用的接收狀態設定爲0，同時將接收 設定爲0傳送到上側的陣列運算單元40。下左右側的情形 亦相同。據此，結束處理器41的1次接收算法。處理器 41是用一定間隔，監視狀態暫存器60的內容，每當上下 117 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） —*—---------------訂--------- (請先閱讀背面之注意事項再填寫本頁) 經濟部智慧財產局員工消費合作社印製 540010 A7 ___ B7 五、發明說明（/Μ) 左右任何的接收狀態爲1時即開始該接收算法。又，即使 處理器41不用一定間隔監視狀態暫存器60的內容，亦可 利用插入處理進行組裝。 又，該陣列運算單元40，主要是根據一個以上的輸入 圖像生成一個輸出圖像爲前提加以說明，但在不同的用途 時需變更電路以在計算途中輸出計算資料。此時，只要變 更程式以僅增加應輸出標記解碼器58的結果送達的計算資 料的數量，將對應結果資料暫存器62所讀入的計算資料的 結果送達設定爲1即可。 以上’說明了本實施形態，但本發明不限於上述實施 形態’若是熟習該領域之業者的話，亦能實施各種的樣態 ’在不脫離本發明的技術思想範圍內，當然可適當改變本 發明的構成，該改變也屬於本發明的技術範圍。 [產業上的可利用性] 根據申請專利範圍第丨及2項之發明，當即使使用明 度、彩度、色相等亦不易從動態圖像中將移動物體單體或 其一部分’作爲一塊的領域切出時，本發明能計算前述移 動物體的數量。例如，當計算活的透明變形蟲(ameba)時， 當然無法對整個變形蟲施以著色。又，即使將光源與背景 設定爲適當的顏色，變形蟲亦爲相同的色，或因光的折射 、反射等使變形蟲的中央與邊緣成爲不同的色，亦不易得 到全面塗抹變形蟲的圖像。大部分的情形中，因光的折射 、反射等’在變形蟲的輪廓部分出現特異的輝度値。因此 ’若使用本發明的話，根據這種特異的輝度値生成邊緣資 118 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） --ί---,------裝----丨！訂--------- (請先閱讀背面之注意事項再填寫本頁) 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明（/^7) 訊，就能從背景切出整個變形蟲，故變形蟲之數量的計算 較爲容易。同樣地，也適用於微塵粉等的微生物、白血球 與精子等細胞。當然，對於能根據蝌蚪與人臉的顏色資訊 容易地區別出背景的移動物體，也能適用本發明。由於在 移動物體、光源及背景中，不必特別做任何處理，即能計 算移動物體數量，故在能拍攝移動物體的現有裝置中連接 本發明’即能廉價地實現移動物體用的物體計數裝置。此 外，亦能在判定動態圖像中有無移動物體的前處理中亦能 加以利用。再者，藉自本發明中直接取出從背景分儺出的 物體領域輸入另外的裝置，以辨識移動物體時的前處理中 亦能加以利用，進而廉價地實現物體辨識裝置。 根據申請專利範圍第3項之發明，當即使使用明度、 彩度、色相等亦不易從靜止圖像或動態圖像的框圖像中， 將物體單體或其一部分作爲一塊的領域切出時，本發明能 計算前述移動物體的數量。例如在計算透明的珠子(beads). 時’當然無法對珠子進行著色。又，即使將光源與背景設 定爲適當的顏色，珠子亦會成爲相同的色，或因光的折射 、反射等，使珠子的中央與邊緣成爲不同的色，皆不易得 到全面塗抹珠子的圖像。大部分的情形中，因光的折射、 反射等在珠子的輪廓部分出現特異的輝度値。因此，當使 用本發明時，由於能根據該特異的輝度値生成邊緣資訊， 據以從背景切出整個珠子，故計算珠子的數量較爲容易。 同樣地，也適用於微塵粉等的微生物、白血球與精子等細 胞。當然，根據蝌蚪與人臉的顏色資訊容易地與背景區別 119 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） —·—*------裝--------訂--------- (請先閱讀背面之注意事項再填寫本頁) 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（/Μ) 的移動物體，亦能利用本發明。如上述般，由於不需對物 體、光源及背景施以任何處理即能計算物體數量，故藉在 能拍攝物體的現有裝置中連接本發明，即能廉價地實現物 體用的物體計數裝置。此外，判定靜止圖像中有無物體的 前處理中，亦能加以利用。再者，藉自本發明中直接取出 從背景分離出的物體領域輸入另外的裝置，以辨識移動物 體時的前處理中亦能加以利用，進而廉價地實現物體辨識 裝置。 根據申請專利範圍第1、2及3項之發明，當即使使用 明度、彩度、色相等亦不易從動態圖像中，切出移動物體 及靜止物體單體或其一部分作爲一塊的領域時，本發明能 計算前述物體中移動物體數與所有物體之數量。例如，當 計算透明的變形蟲(ameba)中活的透明變形蟲與整個變形蟲 時，能將一定時間內移動的變形蟲視爲活的。藉使用動態 圖像中相隔一定時間，的兩個框圖像、或相隔一定時間拍攝 的兩個靜止圖像，本發明能計算移動的變形蟲數與所有的 變形蟲數。同樣的，也適用於微塵粉等的微生物、白血球 與精子等細胞。當然，對根據蝌蚪與人臉的顏色資訊而能 容易地與背景區別的物體，也能利用本發明。如上述般， 由於能以一種裝置計算移動物體數與所有物體數，故能廉 價地貫現計算移動物體數與所有物體數的裝置。此外，判 定靜止圖像中有無物體的前處理中，亦能加以利用。再者 ，藉自本發明中直接取出從背景分離出的物體領域輸入另 外的裝置’以辨識移動物體時的前處理中亦能加以利用， * —«I 1#. ϋ n IV- I I n n ϋ 0 n n 1 · (請先閱讀背面之注意事項再填寫本頁)· I I I I • Line • 丨 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 540010 Α7 Β7 V. Description of the invention (/ ^) The number Γυ acts as a magnification from 0 to 1 (refer to step 1607). As shown in Fig. 31, when the triangular edge information 154 in which the edge information 114 is a broken line is formed, interpolation of the broken line must be performed. First, when the system operates using the edge information 154 of the triangle in the dotted state (see step 1605), although the phase difference between the inside and the outside of the edge information 154 of the triangle in the dotted state is more than 90 degrees, the inside and outside of the triangle The boundary is not clear. Therefore, each AOUu calculates the output λ u of the non-linear oscillator (see step 1609). When this output λ ϋ is 1, the non-linear oscillator when Au is set to 1 in the nearby non-linear oscillators, then the parameters Ψϋ and Ψ kl are more than Θ. That is, 'λ ϋ and λ kl are about the same phase. If Θ is positive 値, the phase difference at the worst time will not exceed 90 degrees. The maximum 値 of this phase difference is determined by Θ 値, in the range where λ ϋ and I u are 1 at the same time, if the value is increased to 0, the phase difference will approach 0 degrees. Therefore, when λϋ and Au are used, the contour parameter 7 / u showing approximately the same phase number in a nearby nonlinear oscillator is calculated according to Equation 48 (refer to step 1610). Next, if the contour parameter 7? Ϋ is about half in the entire vicinity, the boundary parameter of the magnification of the combined 値 is reduced according to Equation 49, and otherwise, it is increased according to Equation 49 (see step 1611). For example, if it is near 8, if it is between 3 and 5, it is sufficient to reduce the boundary parameter according to Equation 49. If the process is repeated for continuous operation, when the triangle edge information 154 of the dotted state shown in FIG. 31 is obtained, the inner region 155 of the dotted triangle and the outer region 156 of the dotted triangle are separated. As shown in Fig. 32, when two triangles are overlapped, edge information 157 of the front triangle and edge information 158 of the rear triangle can be obtained. At this time, the 109 paper size is applicable to the Chinese National Standard (CNS) A4 specification (210 x 297 public love) (please read the precautions on the back before filling this page)-binding-order --------- line Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 540010 Α7 Β7 V. Description of the invention (/ θ) · Using 3 areas of the inner area 159 of the front triangle, the inner area 160 of the rear triangle, and the background area 161 of the double triangle The phase deviations of the non-linear oscillators are separated into three areas. Also, as shown in FIG. 33, even if the edge information 162 of two overlapping circles is a dotted line, it is separated into three areas of the front circular inner area 163 and the rear circular inner area 164 and the double circular background area 165. Areas. Therefore, using the data processing device 110 constituted by the array operation unit 40 and the visual device corresponding to the 12th patent application scope of the object / background separation mechanism 16 can be formed with the edge information 114 forming the edge information image 115 , Separate the object area 141 from the background. In the above, the visual devices of the 6th patent application to the 12th patent application have been described. Of course, although these visual devices can be assembled using a general computer, when the moving object 2 is counted, each of the aforementioned mechanisms must be executed at high speed according to the moving speed of the moving object 2. In particular, when the image size or resolution of the frame image 1 is increased, the image memory mechanism 12 (see FIGS. 1 and 6) and the image vibration mechanism 13 (see FIG. 3) using the image itself as a processing target. , Figures 4 and 5), edge information generating mechanism 14 (refer to Figures 1 and 6), edge information generating mechanism 15 (refer to Figures 1 to 5 and 7 to 12), object / background separation Mechanism 16 (refer to Figures 2, 4, 5, 7 to 12), position / size detection mechanism 17 (refer to Figures 1 and 6), domain standardization mechanism 27 (refer to Figure 7), normalized image The holding mechanism 28 (refer to FIG. 7) and the image recognition mechanism 29 (refer to FIG. 8) increase the calculation amount for each width direction and each height direction, that is, increase in proportion to the image size or resolution. Therefore, the scope of the patent application ranges from 6 to 110. The paper size is applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm). --Order --------- (Please read the notes on the back before filling this page) Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 540010 A7 ____ B7 V. Description of the invention, please refer to the 12th in the scope of patents In some cases, the visual device may not achieve the expected performance due to different uses. Here, in order to use digital technology to assemble the image memory mechanism 12, the image vibration mechanism 13, the edge information generation mechanism 14, the edge information formation mechanism 15, the object / background separation mechanism 16, the position / size detection mechanism 17, and the field normalization mechanism 27. And normalized image holding mechanism 28. The array operation unit 40 in the thirteenth patent application is arranged in the data processing device 110 in a grid shape as shown in FIG. 13, and the array operation unit 40 is The wiring is provided so as to be able to communicate only with the adjacent array operation unit 40 in the data processing device 110. That is, direct wiring is carried out between each other in the vicinity of 4. Therefore, compared with the case of wiring between 8 units, it can operate at the same speed with less electronic parts and wiring, and it can be easily expanded in the future when it is necessary to expand the nearby size. The array operation unit 40 of the scope of application for patent No. 14 is shown in FIG. 34, which is composed of a processor 41 used to calculate a pattern processing formula, and used to memorize all parameters, constants, and functions used in the formula. It is composed of a memory (Memory) 42 of an operator, and a controller (Controller) 43 for communicating with a nearby array operation unit 40. The processor 41 may use the address specified by the address bus 51. Address, select any memory element and register of the memory 42 and the controller 43. In addition, the processor 41 is connected to the memory 42 and the controller 43 through the address bus 51 to enable bidirectional communication to access data (DATA) of any memory element and register specified by the address bus 51 . When the array operation unit 40 inputs a Front Input Data Set consisting of more than one input pixel, 111 paper sizes are applicable to the Chinese National Standard (CNS) A4 specification (210 x 297 mm) I- --------- Installation -------- Order --------- ^ 9— (Please read the notes on the back before filling this page) 540010 A7 B7 V. Invention Explanation (丨 0) The controller 43 stores the previously input data group in the memory 42. In addition, the controller 43 transmits the calculation data in the memory 42 made by the function to the adjacent array operation unit 40, and stores the calculation data received from the adjacent array operation unit 40 in the memory 42. If necessary, it is transmitted to the array operation unit 40 other than the input. Finally, the controller 43 outputs the output image data as result data. As described above, the reason why the controller 43 is mounted in the array operation unit 40 is because the processor 41 can also operate during the communication between the array operation units 40, so the processor 41 can also perform calculations during the communication waiting time. High-speed processing, and even if the number of array operation units 40 in the vicinity is changed, there is no need to change hardware, and the controller 43 performs edge processing of the image, that is, it can automatically perform exception processing on edge pixels in the image, so the processing The program of the processor 41 is extremely simple without having to perform edge processing. Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs ----- * ------ • Installation -------- Order (Please read the precautions on the back before filling this page) Processor 41 and The memory 42 can use a general digital circuit. The specific circuit diagram of the controller 43. is shown in FIG. The Address Buffer (53) receives the address from the processor (41) through the Address Bus (5: 1). 'According to the Address Decoder (54), each register is selected. And other functional blocks. The data buffer (Data Buffer) 55 receives the data from the processor 41 via the Data Bus 52, and through the register and internal data selected by the address decoder 54 The bus 56 performs exclusive communication. The communication direction is specified by Read. When the address designation flag register (Flag Reg1Ste: r) 57 is used, the data is stored in the tag 57 register and used. Flag Decoder 58 decodes and sends complex signals (Signals) to adjacent array operations. 112 This paper is sized for China National Standard (CNS) A4 (210 X 297 mm) 540010 A7 B7___ 5 Description of the invention (/ (/) (Please read the notes on the back before filling in this page) Yuan 40. The complex signal is received using the flag encoder (Flag Encoder 〇59, after analysis, it is stored in the state register ( Status Register) 60, RECIEVER) returns to the array operation unit 40 of the transmission source. The reception is received by the marker encoder 59 of the transmission source of the complex signal. As a result, it is confirmed that the transmission of the complex signal is completed. When the status register is selected by the address, the status The content of the register 60 is transmitted to the processor 41 as data through the data bus 52. When the mark encoder 59 receives one or more front input inputs corresponding to one or more input images (Front Input Sent) , The front input data set (Front Input Data Set) composed of more than one input image is read into the front input data register (Front Input Data Register) 61 with the required memory capacity points. When using the bit When the front input data register 61 is selected by the address, the content of the front input data register 61 is transmitted to the processor 41 as data. After the processor 41 finishes the calculation, the result data register (Result Data) is selected according to the address. Register) 62, the result data register 62 reads the image data of the output image as the result data (Result Data). At the same time, the mark encoder 59 transmits the result to (Resul t Send). The Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs will print the data required for calculation from the nearby array computing unit 40, and then select the Output Data Register 63 as the address. The data transmitted to the nearby array operation unit 40 is used as calculation data and is read into the output data register 63. After that, it is transmitted to all adjacent array operation units 40 as calculation data. When the complex signal (Signals) is received from the array operation unit 40 on the upper side, the calculation data is read into the Upper Input Data Register 64. After that, 113 paper sizes are applicable to the Chinese National Standard (CNS) A4 (210 X 297 mm). Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs. 540010 A7 B7. 5. Description of the invention (/ u) Input information according to the address selection After the register 64, the content of the input data register 64 will be transmitted as calculation data. The same applies when receiving plural signals from the array operation unit 40 on the lower side, the left side, and the right side. The lower input data register 65, the left input data register 66, and the right input stock register 67 all operate in the same manner. The blocks of the various buffers, various registers, and the address decoder 54 are general electronic circuits. Specifically, the tag decoder 58 and the tag encoder 59 have input and output signals shown in Figs. 36 and 37. Type (Type): 5 types of contents read from the data register (〇utpUt Data Register) 63. This number of bits is sufficient to distinguish all calculation data transmitted and received by the array operation unit 40. Counter 1 X (Count-X) and Counter 1 Y (Count-Y) show each integer without 4-bit code, and show the number of transfers between the array operation unit 40. When the array operation unit 40 transmits calculation data, each counter is 0. When the calculation data transmitted from the left and right array operation units 40 is transmitted again, 値 is incremented by 1 in the counter -X of the mark encoder 59. When transmitting the calculation data transmitted from the upper and lower array operation unit 40, 値 is incremented by 1 in the counter-Y of the mark encoder 59. The processor 41 is in the Send Flag of the mark register 57 and specifies that the content of the output data register 63 is transmitted to any of the upper, lower, left, and right directions. When the mark decoder 58 receives the specified output data temporarily In the central decoding of the address decoder 54 of the memory 63, the mark decoder 58 is about to send and output it in accordance with the specified direction of the mark. The delivery mark is displayed in 4 bits. When the calculation data of the array operation unit 40 is transmitted to the Quartet's array operation order 114 $, the paper size is applicable to the Chinese National Standard (CNS) A4 specification (210 X 297). I ----- ----- 丨 ^^ Equipment ---- I I--Order --------- (Please read the notes on the back before filling out this page) Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 540010 A7 B7 V. Description of the invention (//)). At 40 yuan, the processor 41 is set to 1111 'When the calculation data transmitted from the array operation unit 40 on the right is transmitted to the upper, lower and left sides, it is set to 1101, when transmitting from the lower side to the upper side, is set to 0100 when transmitting from the upper side to the lower side. In this way, since not only can be transmitted efficiently without repeated transmission, but also the rules for determining the transmission direction become clear, the mark encoder 59 can determine which array it is from by combining the type, counter-X and counter-Y Calculation unit 40, which kind of calculation data is transmitted. When the calculation data is read into the result data register 62 as the result data, the mark decoder 58 also receives the result decoding (Result Decoding), and sends the result to the (Result Send). When the marker encoder 59 receives the delivery from any of the four directions, that is, the type of the receiving direction and the counter-X and the counter-Y are updated, and the content of the state register 60 of the part is updated. At the same time as this update, set the reception to 1 to the reception direction. The tag encoder 5 · 9 of the array operation unit 40 of the transmission source performs reception at the instant when the reception becomes 1, and updates the reception status (Receive Status) of the status register 60. Therefore, in each array operation unit 40, the processor 41 only checks the receiving state of the state register 60, that is, it can determine which input data register is stored in the valid calculation data. Therefore, for example, if the calculation data is read into the upper input data register 64, although the processor 41 can read the data from the upper input data register 64 by a specified address, at the same time, the Upper Decoding is read from The address decoder 54 is transmitted to the mark encoder 59, and the upper part returns to 0 in the reception state, and the reception on the upper side is transmitted as 0. The same applies to the lower left and right sides. When the mark encoder 59 receives only one input image, the 115 paper size is applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm). ---- (Please read the notes on the back before filling this page) Order -------- 540010 A7 B7 V. Description of the invention (/ φ (Please read the notes on the back before filling this page) Input delivery, that is, the front input send status for the input image corresponding to the input input delivered before receiving in the state register 60 is 1. Also, when the processor 41 temporarily removes the front input data for the input image, When the memory 61 reads the data, the address decoder 54 transmits the Front Decoding to the mark encoder 59, and sets the corresponding front input delivery status corresponding to the received front input delivery to 0. The processor 41 borrows and reads Take the content of the state register 60, that is, determine whether the latest input image is memorized in the previous input data register 61. The print processor 41 of the employee consumer cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs transmits the calculation data to the controller 43 The algorithm of the square array operation unit 40 is shown in FIG. 38. Fig. 38 shows a mixed process of program control using the processor 41 and hardware logic using the tag decoder 58 and the tag encoder 59. In step 38 of Fig. 38, the processor 41 reads the status register 60 In step 72, it is judged whether the receiving status in the read content is all 0. If NO, the processing is ended, and if YES, the processing moves to step 73. In step 73, the processor 41 decides to transmit to the phase. The data type, counter, and transmission direction of the adjacent array operation unit 40 are written into the tag register 57. In step 73, the processor 41 writes the data transmitted by the adjacent array operation unit 40 into the output data. The register 63. In step 75, the content of the output data register 63 is used as calculation data and transmitted to the adjacent array operation unit 40. In step 76, only the delivery of the tag register 57 to the tag office is performed. In the specified direction, the delivery is set to 1 for transmission. By this, one transmission of the processor 41 ends. The processor 41 starts the transmission algorithm when the data to be transmitted is updated in the memory 42 each time. . 116 This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm) 540010 Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7 V. Description of the invention (/ (Γ) Controller 43 is calculated from the array on the upper side The algorithm when the unit 40 receives the calculation data is shown in Fig. 39. Fig. 39 shows the processing of the hardware logic using the tag decoder 58 and the tag encoder 59. In Fig. 39, step 81 is the input of the tag encoder 59 Delivery. In step 82, the tag encoder 59 determines whether the delivery is 1. If it is NO, the process is ended, and if YES, the process moves to step 83. In step 83, the upper input data register 64 reads the calculated data transmitted from the upper side. In step 84, the mark encoder 59 sets the reception status for the upper side in the status register 60 to 1, and transmits the reception setting to 1 to the array operation unit 40 on the upper side. The same is true for the upper and lower right. With this, the one-time reception algorithm of the controller 43 ends. The controller 43 monitors the delivery from the up, down, left, right, and left array operation units 40 at any time, and starts the receiving algorithm each time the delivery is received. The algorithm when the processor 41 receives data from the input data register 64 is shown in FIG. Fig. 40 shows a mixed process of program control using the processor 41 and hardware logic using the tag decoder 58 and the tag encoder 59. In Fig. 40, in step 91, the processor 41 reads the contents of the state register 60. In step 92, it is determined whether or not the received status for the upper side of the read content is 1. If it is NO, the process is ended, and if it is YES, the process proceeds to step 93. In step 93, the processor 41 reads data from the input data register 64. In step 94, the marker encoder 59 sets the reception status for the upper side in the status register 60 to 0, and transmits the reception setting to 0 to the array operation unit 40 on the upper side. The same applies to the lower left and right sides. As a result, the primary reception algorithm of the processor 41 is terminated. The processor 41 monitors the contents of the status register 60 at regular intervals. Whenever the upper and lower 117 paper sizes are applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) — * —------- -------- Order --------- (Please read the notes on the back before filling out this page) Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economy 540010 A7 ___ B7 V. Description of the invention ( / M) The receiving algorithm starts when any of the receiving statuses is left and right. In addition, even if the processor 41 does not need to monitor the contents of the state register 60 at regular intervals, it can be assembled by insert processing. The array operation unit 40 is mainly explained on the premise that one output image is generated based on one or more input images, but the circuit needs to be changed for different applications to output calculation data during calculation. At this time, as long as the program is changed to increase only the amount of calculation data delivered by the result of the mark decoder 58, the result delivery of the calculation data corresponding to the result data register 62 is set to 1. As described above, the present embodiment is described, but the present invention is not limited to the above embodiments. If a person skilled in the field can implement various aspects, the present invention can be appropriately changed without departing from the technical scope of the present invention. This configuration also belongs to the technical scope of the present invention. [Industrial Applicability] According to the inventions in the scope of patent applications No. 丨 and No. 2, even when the brightness, chroma, and color are equal, it is not easy to take a single moving object or a part of it as a piece from a moving image When cutting out, the present invention can calculate the number of the aforementioned moving objects. For example, when calculating live amoeba, it is of course impossible to color the entire amoeba. In addition, even if the light source and the background are set to appropriate colors, the amoeba is the same color, or the center and edge of the amoeba are different colors due to light refraction, reflection, etc., and it is not easy to obtain a picture of the entire amoeba. image. In most cases, due to the refraction, reflection, etc. of light, a specific brightness 値 appears in the outline portion of the amoeba. Therefore, if the present invention is used, 118 edge papers will be generated based on this specific luminance. This paper size is applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) --ί ---, ----- -Equipped ---- 丨! Order --------- (Please read the notes on the back before filling this page) Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 540010 A7 B7 V. Description of Invention (/ ^ 7) The entire amoeba is cut out from the background, so the calculation of the number of amoeba is relatively easy. The same applies to microorganisms such as dust particles, cells such as white blood cells and sperm. Of course, the present invention can also be applied to moving objects that can easily distinguish the background based on the color information of the face and the face. Since the number of moving objects can be counted in the moving object, the light source and the background without any special treatment, the present invention can be connected to an existing device capable of photographing a moving object to realize an object counting device for moving objects at low cost. In addition, it can also be used in pre-processing to determine the presence or absence of a moving object in a moving image. Furthermore, by taking out the object field separated from the background directly from the present invention and inputting another device, it can also be used in the pre-processing when recognizing a moving object, thereby realizing an object recognition device at low cost. According to the invention in item 3 of the scope of patent application, when it is not easy to cut out a single object or a part of it as a block from a frame image of a still image or a moving image even if the brightness, chroma, and color are equal. The present invention can calculate the number of the aforementioned moving objects. For example, in the calculation of transparent beads. Of course, the beads cannot be colored. In addition, even if the light source and the background are set to appropriate colors, the beads will become the same color, or the center and edge of the beads will be different colors due to light refraction and reflection, etc. . In most cases, due to the refraction and reflection of light, a specific brightness 値 appears in the outline of the beads. Therefore, when the present invention is used, since the edge information can be generated based on the specific luminance 値, and the entire beads can be cut out from the background, it is easy to calculate the number of beads. The same applies to microorganisms such as dust particles, cells such as white blood cells and sperm. Of course, according to the color information of the face and the face, it is easy to distinguish from the background. 119 This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 mm). ---- Order --------- (Please read the notes on the back before filling out this page) 540010 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7 V. Movement of the Invention Description (/ Μ) Objects can also make use of the invention. As described above, since the number of objects can be counted without any treatment of the objects, light sources, and backgrounds, by connecting the present invention to an existing device capable of photographing objects, an object counting device for objects can be realized at a low cost. It can also be used in preprocessing to determine the presence or absence of an object in a still image. Furthermore, the object area separated from the background can be directly taken out from the present invention and inputted into another device, which can be used in the pre-processing for identifying a moving object, thereby realizing an object identification device at low cost. According to the inventions in claims 1, 2, and 3 of the scope of patent application, when it is not easy to cut out a moving object or a stationary object or a part of it as a piece from a moving image even if the brightness, chroma, and color are equal, The invention can calculate the number of moving objects and the number of all objects among the aforementioned objects. For example, when calculating the live transparent amoeba and the entire amoeba among the transparent amoeba, the amoeba that moves within a certain period of time can be regarded as alive. By using two frame images separated by a certain time in a dynamic image, or two still images taken by a certain time, the present invention can calculate the number of moving amoeba and all the amoeba. The same applies to microorganisms such as dust particles, cells such as white blood cells and sperm. Of course, the present invention can also be used for objects that can be easily distinguished from the background based on the color information of the face and the face. As described above, since the number of moving objects and the total number of objects can be calculated by one device, a device for calculating the number of moving objects and the total number of objects can be implemented inexpensively. It can also be used in preprocessing to determine the presence or absence of an object in a still image. In addition, the present invention can also be used in the pre-processing when recognizing a moving object by directly taking out the object area separated from the background and inputting another device. * — «I 1 #. Ϋ n IV- II nn ϋ 0 nn 1 · (Please read the notes on the back before filling this page)
ϋ_1 mmemmm em— 1*-m. I mmmam ϋ ·ϋ i —m n I 着· 120 540010 A7 B7 五、發明說明（/^7) 進而廉價地實現物體辨識裝置。 (請先閱讀背面之注意事項再填寫本頁) 根據申請專利範圍第1、2及3項之發明，當即使使用 明度、彩度、色相等亦不易或無法從動態圖像中切出移動 物體及靜止物體單體、或將其一部分作爲一塊的領域時， 本發明能計算前述物體中移動物體數與所有物體數，計算 移動物體數與靜止物體數或其比例。例如，當求透明的變 形蟲的生存率時，能將某一定時間內移動的變形蟲視爲活 的。藉使用動態圖像中相隔一定時間的兩個框圖像、或相 隔一定時間拍攝的兩個靜止圖像，本發明能計算移動的變 形蟲數與所有的變形蟲數，故能簡單地求出變形蟲的生存 率。同樣的，也適用於微塵粉等的微生物、白血球與精子 等細胞。當然，對根據蝌蚪與人臉的顏色資訊而能容易地 與背景區別的物體，也能利用本發明。如上述般，由於能 用一種裝置計算移動物體的數量與靜止物體的數量，故能 廉價地實現計算移動物體與靜止物體之比例的裝置。其他 ，在動態圖像中，也能利用在判定有無移動物體或靜止物 體的前處理。再者，藉自本發明中直接取出從背景分離出 的物體領域輸入另外的裝置’以辨識移動物體或靜止物體 經濟部智慧財產局員工消費合作社印製 時的前處理中亦能加以利用’能廉價地實現物體辨識裝置 〇 根據申請專利範圍第4及5項之發明’使用由三原色 波長、可視光波長、紅外線波長’紫外線波長、及其他所 有的電磁波中由任意的頻帶構成的動態圖像’本發明能搜 索攝像機能拍攝範圍的物體。當物體移動時’能藉檢測移 121 本紙張尺度適用中國國家標準（CNS)A4規格（21〇 x 297公爱） 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明（/>) , 動物體的位置及大小，使移動式攝像機的方向對準物體， 調整倍率以使圖像中的移動物體大小爲一定。又，當物體 靜止時，由於能使移動式攝像機微細地振動據以檢測靜止 物體的位置與大小，故對靜止物體亦能與移動物體時相同 的控制移動式攝像機。再者，由於亦能從外部控制移動式 攝像機，故能與電腦等組合加以利用。因此，本發明能利 用如下。例如，除了在走廊等照明較穩定的場所外，在玄 關與屋外等雜訊較多的地方進行人、車或行李等物體之監 視時，本發明能控制移動式攝像機以用適當的倍率拍攝該 等物體。另外，在便利商店或超商等動態較少的地方藉集 中抬攝顧客，本發明能達到防止偷竊等效果。如上述般， 由於能用一個裝置對廣範圍的物體進行監視，故能廉價地 實現物體監視裝置。除此之外，在判定動態圖像中有無特 定物體的前處理中亦能加以利用。再者，藉自本發明中直 接取出從背景分離出的物體領域加以規格化之圖像輸入另 外的裝置，以辨識移動物體或靜止物體時的前處理中亦能 加以利用，能廉價地實現物體辨識裝置。 又，根據申請專利範圍第5項之發明，使用由三原色 波長、可視光波長、紅外線波長，紫外線波長、及其他所 有的電磁波中由任意的頻帶構成的動態圖像，本發明能搜 索攝像機能拍攝範圍的物體。將該等物體以適當的辨識方 法分類爲若干種類後，與位置或時刻等之屬性同時加以記 錄。因此，能計算一定時間內位於移動式攝像機能拍攝範 圍之特定物體的數量，或記錄物體移動的軌跡。例如，在 122 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） —----*------^-------It--- (請先閱讀背面之注意事項再填寫本頁) 考 540010 A7 B7 五、發明說明（/w) (請先閱讀背面之注意事項再填寫本頁) 經濟部智慧財產局員工消費合作社印製 卡拉OK包廂與商店街等地方，本發明能計測顧客數量與 通行量等。另外，在便利商店、超商及工場等處，能調查 顧客的聚集場所或記錄作業員的作業步驟。另外，在遊樂 園場等處，能使用於「不倒翁倒下」的移動物體搜索遊戲 與或射撃遊戲等。更進一步的，藉將本發明組裝於機器人 中，工業用機器人即能搜索製造線中的工件、或偵測人員 之接近而使動作停止，是以本發明能實現人型機器人及導 盲犬機器人等移動機器人的視覺裝置。因此，若在汽車、 公共汽車、砂石車等車輛的前後左右安裝本發明的話，本 發明能發現位於駕駛員死角的物體進而發出警報，亦能預 先偵測接近而來之車輛等移動物體，也能取得發生交通事 故時的記錄。對坐在輪椅的人而言，由於不必轉頭就能擴 大視野，故能預先察知由後方逼近的自行車或汽車，能有 效防止事故。此外，若在電車等之車門口設置本發明，本 發明也能發現被門夾住或站在門旁的乘客。若車站的月台 設置本發明的話，本發明能發現掉落於軌道上的物品，或 對站在月台之軌道邊的人發出警告。另外，若將本發明設 置在救難用機器人的話，救難用機器人即能找出在海岸、 海上或河川等處的遇難者，自動地接近以將救生衣或繩索 等搬送給遇難者，或將遇難者搬送到安全的地點。該救難 用機益人亦g旨利用於搜索因地震而被埋在瓦碟下的人。如 上述般，由於通用一^個裝置進ί了物體的搜索、計數與移動 軌跡的記錄，故能廉價地實現物體搜索裝置、物體辨識裝 置、物體計數裝置及物體記錄裝置等。 123 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） 經濟部智慧財產局員工消費合作社印製 540010 A7 B7 五、發明說明（/U) ^ 根據申請專利範圍第6項之發明，能在並列輸入數位 圖像的各像素，以圖像單位或像素單位使數位圖像振動後 ，並列輸出數位圖像的各像素。由於本發明不使用振動台 等物理性機構，故能高速地使數位圖像振動。因此，在移 動物體係以高速移動之情形中，必須即時生成數位圖像中 的所有物體的邊緣資訊時，本發明能達成所需之處理速度 〇 根據申請專利第7項範圍之發明，能在並列輸入數位 圖像的各圖像，以生成數位圖像的各像素的粗邊緣資訊後 ，並列輸出粗邊緣資訊。所生成的粗邊緣資訊因受物體的 移動方向、移動速度及物體與背景的顏色資訊差等的影響 ，雖未必正確，但由於在有太陽光照射的屋外等雜訊較多 的環境亦不必特別對數位圖像進行修正，而能生成任意形 狀的移動物體的粗邊緣資訊，因此本發明能不受使用環境 的限制加以使用。又，本發明對紅外線、紫外線、甚至於 放射線等能與可視光波長、三原色波長同樣地加以處理， 且越增加頻帶數越能減少雜訊的影響。此時，由於本發明 不必變更記憶容量以外之資料處理裝置的電路，故能容易 地增加頻帶數。因此，對不僅在屋內且能在屋外活動的機 器人來說，本發明能高速且便宜地提供移動物體的輪廓、 位置及大小。此外，本發明對不限定對象物的視覺辨識裝 置之前處理而言，亦能有效地動作。 根據申請專利範圍第8項之發明，能在並列輸入粗邊 緣資訊及數位圖像的各圖像，利用數位圖像從粗邊緣資訊 124 玉紙張尺度適用中國國家標準（CNS)A4規格（21〇χ 297公釐) ^ -I I n ϋ ·ϋ ·ϋ n n ϋ · n I 1 n ϋ i.— I 一-01 1 n I (請先閱讀背面之注意事項再填寫本頁) 着· 540010 經濟部智慧財產局員工消費合作社印製 A7 B7_ 五、發明說明（Λ]) 生成形成邊緣資訊後，並列輸出形成邊緣資訊。由於本發 明能將以任意方法生成的粗邊緣資訊形成爲更明瞭且明確 之邊緣資訊，因此能容易地減低生成粗邊緣資訊之機構所 需要的負載。此外，由於本發明不太受粗邊緣資訊的影響 ，因能將數位圖像設定爲低解像度的低解像度數位圖像的 特定領域所生成的粗邊緣資訊，形成爲更明瞭且明確之邊 緣資訊。因此，不需增加硬體量與計算量，本發明能對低 倍率數位圖像中的物體生成高精度的邊緣資訊。藉本發明 ，能使過去以廣角攝像機與高精細攝像機的物體辨識裝置 ，用一個攝像機實現物體辨識。 根據申請專利範圍第9項之發明，能在並列輸入邊緣 資訊，檢測用邊緣資訊所顯示的物體的位置及大小後，將 物體的位置及大小作爲重複資訊加以並列輸出。本發明， 能以需要大量的硬體量與計算量所檢測的圖像中之複數個 物體的位置及大小相同程度之品質或之上的品質，…次性 的從複數個物體的邊緣資訊中檢測出位置及大小。從用攝 錄影機所拍攝的動態圖像的框圖像或用數位相機所拍攝的 靜止圖像中’檢測該圖像中的複數物體之位置及大小的前 處理中亦能利用本發明，以高速且廉價地實現靜止圖像及 動態圖像的圖形辨識算法。 又，根據申請專利範圍第9項之發明，能在並列輸入 物體領域圖像的各像素，檢測用物體領域圖像的各像素所 顯示的物體的位置及大小後，將物體的位置及大小作爲重 複資訊加以並列輸出。本發明，能以需要大量的硬體量與 125 I紙張尺度適用中國國家標準（CNS)A4規格（21G X 297公爱) " 一— 1-----------裝--------訂--------- (請先閱讀背面之注意事項再填寫本頁) 540010 經濟部智慧財產局員工消費合作社印製 Α7 Β7 五、發明說明（/γ) / 計算量所檢測的圖像中之複數個物體的位置及大小相同程 度之品質或之上的品質，一次性的從複數個物體的邊緣資 訊中檢測出位置及大小。尤其是當檢測像細胞或精子般具 有圓形部分的物體，或像人臉般可視爲圓形的物體位置及 大小時，由於不需從該等物體領域再度生成邊緣資訊，故 能高速且廉價地實現靜止圖像及動態圖像的圖形辨識算法 〇 根據申請專利範圍第10項之發明，能在並列輸入物體 領域圖像的各像素及資料圖像的各像素，以將數位圖像中 的物體領域規格化後，並列輸出規格化圖像的各像素。本 發明，能以需大量的硬體量與計算量所生成、對位置偏離 較弱'多間隙的規格化圖像同程度之品質或之上的品質， 使用物體領域圖像的物體領域以外，將被遮蔽的數位圖像 的物體領域配合數位圖像的尺寸塡補間隙，同時進行規格 化。對用攝錄放機拍攝的動態圖像之框圖像或用數位相機 拍攝的靜止圖像中切出的特定物體進行規格化的前處理中 亦能加以利用，而能高速且廉價地實現靜止圖像及動態圖 像的圖形辨識算法等。 根據申請專利範圍第11項之發明，能在並列輸入若干 模板圖像的各像素，並列輸入規格化圖像的各像素，進行 模板圖像與圖形匹配後，並列輸出匹配結果圖像的各像素 。若對不同位置與大小的相同物體之規格化圖像的再現性 高的話，本發明能選擇僅用附近處理、類似於規格化圖像 的若干模板圖像。是以，由於本發明能將最小平方誤差與 126 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） —-—f------裝--------訂---------^9. (請先閱讀背面之注意事項再填寫本頁) 540010 經濟部智慧財產局員工消費合作社印製 A7 B7 五、發明說明（Λ^) 〆 中性網路等的大區域處理抑制在最小限度’故能高速且廉 價地實現靜止圖像及動態圖像的圖形辨識算法等。 根據申請專利範圍第12項之發明，能在並列輸入形成 邊緣資訊，使用非線性振盪器將物體領域與背景領域分離 後，並列輸出物體領域圖像的各像素。奔發明，針對數位 圖像，不需預先進行邊緣資訊的生成以外之任何前處理， 不受數位圖像中的物體形狀、位置、方向的影響，又，即 使邊緣資訊爲虛線、或交差亦無所謂，再者，即使數位圖 像中的物體領域重疊，亦能分離物體領域與背景領域。另 外，本發明使用數位技術容易組裝硬體，亦能進行適合即 時圖像處理的高速化。 根據申請專利範圍第13及14項之發明，能使視覺裝 置高速地動作。特別是由於並行性的關係，即使圖像尺寸 變大’處理時間亦不變。是以，可利用視覺裝置於需要即 時性之用途。又，由於只要變更程式，即能用同一晶片實 現申請專利範圍第6到申請專利範圍第12項之複數的機構 ，因此本發明能便宜地製造視覺裝置。再者，由於能用同 一晶片實現申請專利範圍第6到申請專利範圍第12項之複 數的機構’因此能將該晶片容易地組裝於移動式攝像機中 ，而增加方便性。最後，由於本發明全部用像素單位輸入 、輸出信號，故藉重疊複數之晶片即能抑制配線量。因此 ，藉三次元VLSI等技術革新，本發明能容易地提升處理性 127 本紙張尺度適用中國國家標準（CNS)A4規格（210 X 297公釐） (請先閱讀背面之注意事項再填寫本頁) 裝 着！ϋ_1 mmemmm em— 1 * -m. I mmmam ϋ · ϋ i —m n I by 120 540010 A7 B7 V. Description of the invention (/ ^ 7) Then implement the object recognition device at low cost. (Please read the precautions on the back before filling out this page) According to the inventions in claims 1, 2 and 3 of the scope of patent application, it is not easy or impossible to cut out moving objects from moving images even if the brightness, chroma, and color are equal. When the area is a single stationary object or a part of it is a block, the present invention can calculate the number of moving objects and the total number of objects among the foregoing objects, and calculate the number of moving objects and the number of stationary objects or a ratio thereof. For example, when the survival rate of a transparent amoeba is obtained, the amoeba that moves within a certain period of time can be regarded as alive. By using two frame images separated by a certain time in a dynamic image or two still images taken by a certain period of time, the present invention can calculate the number of moving amoeba and all the amoeba, so it can be easily obtained Survival rate of amoeba. The same applies to microorganisms such as dust particles, cells such as white blood cells and sperm. Of course, the present invention can also be used for objects that can be easily distinguished from the background based on the color information of the face and the face. As described above, since the number of moving objects and the number of stationary objects can be calculated by one device, a device for calculating the ratio of the moving object to the stationary object can be implemented at a low cost. In addition, in moving images, preprocessing can be used to determine the presence or absence of moving or stationary objects. Furthermore, by taking out the object field separated from the background directly from the present invention and inputting another device 'to identify moving objects or stationary objects, it can also be used in the pre-processing at the time of printing by the employee's consumer cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs'. Realize the object recognition device inexpensively. According to inventions Nos. 4 and 5 of the scope of patent application, 'use a dynamic image composed of an arbitrary frequency band among three primary wavelengths, visible light wavelengths, infrared wavelengths, ultraviolet wavelengths, and all other electromagnetic waves.' The invention can search the objects that the camera can shoot. When the object moves, the paper size can be detected by 121. This paper size is applicable to the Chinese National Standard (CNS) A4 specification (21 × 297). Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. 540010 A7 B7. 5. Description of the invention (/ >), The position and size of the animal body, make the direction of the mobile camera point at the object, and adjust the magnification so that the size of the moving object in the image is constant. In addition, when the object is stationary, the mobile camera can be finely vibrated to detect the position and size of the stationary object, so the stationary camera can be controlled in the same way as when the object is moving. Furthermore, since the mobile camera can also be controlled from the outside, it can be used in combination with a computer or the like. Therefore, the present invention can be used as follows. For example, in addition to places with stable lighting such as corridors, when monitoring objects such as people, cars, or luggage in places with noisy places such as entrances and outdoor areas, the present invention can control a mobile camera to take pictures at an appropriate magnification. And other objects. In addition, the present invention can achieve the effects of preventing theft and the like by taking pictures of customers in a borrowed place in a place with less dynamics such as convenience stores or supermarkets. As described above, since a wide range of objects can be monitored with one device, an object monitoring device can be implemented at low cost. In addition, it can be used in pre-processing to determine the presence or absence of a specific object in a moving image. Furthermore, by taking out the normalized image of the object area separated from the background in the present invention and inputting it into another device, it can also be used in the pre-processing when recognizing a moving object or a stationary object, and the object can be realized inexpensively. Identify the device. In addition, according to the invention in claim 5 of the scope of patent application, the present invention can use a dynamic image composed of an arbitrary frequency band among three primary wavelengths, visible light wavelengths, infrared wavelengths, ultraviolet wavelengths, and all other electromagnetic waves. Range of objects. After classifying these objects into several types by appropriate identification methods, they are recorded at the same time as attributes such as location or time. Therefore, it is possible to calculate the number of specific objects within a certain period of time that the mobile camera can capture, or to record the trajectory of the object movement. For example, in the 122 paper size, the Chinese National Standard (CNS) A4 specification (210 X 297 mm) is applied —---- * ------ ^ ------- It --- (please first Read the notes on the back and fill in this page) Exam 540010 A7 B7 V. Description of the invention (/ w) (Please read the notes on the back before filling in this page) Printed Karaoke Boxes and Stores by the Employee Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs In places such as streets, the present invention can measure the number of customers and the amount of traffic. In addition, in convenience stores, supermarkets, and workshops, it is possible to investigate where customers gather, or to record the steps of workers. In addition, in amusement parks and other places, it can be used for mobile object search games and "shooting games" such as "fall of the tumbler". Furthermore, by assembling the present invention in a robot, an industrial robot can search for a workpiece in a manufacturing line or detect the approach of a person to stop the action. The invention can realize a humanoid robot and a guide dog robot Vision devices for mobile robots. Therefore, if the present invention is installed on the front, back, left, and right sides of vehicles such as cars, buses, and gravel vehicles, the present invention can detect objects located at the driver's dead end and issue an alarm, and can also detect moving objects such as approaching vehicles in advance. Records of traffic accidents can also be obtained. For a person in a wheelchair, because they can widen their field of vision without having to turn their heads, they can know in advance the bicycle or car approaching from the rear, which can effectively prevent accidents. In addition, if the present invention is provided at a doorway of a tram or the like, the present invention can also find passengers caught by the door or standing by the door. If the present invention is provided on a platform of a station, the present invention can detect items falling on the track, or issue a warning to a person standing beside the track of the platform. In addition, if the present invention is installed in a rescue robot, the rescue robot can find the victims on the coast, the sea, or the river, etc., and automatically approach to carry the life jacket or rope to the victims, or send the victims Move to a safe place. The rescue machine is also intended to be used to search for people buried under a tile by the earthquake. As described above, since one device is generally used to record the search, counting, and movement trajectory of an object, an object search device, an object recognition device, an object counting device, an object recording device, and the like can be implemented at low cost. 123 This paper size applies to China National Standard (CNS) A4 (210 X 297 mm) Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 540010 A7 B7 V. Description of the invention (/ U) ^ According to item 6 of the scope of patent application According to the invention, each pixel of a digital image can be input in parallel, and the digital image can be vibrated in image units or pixel units, and then each pixel of the digital image can be output in parallel. Since the present invention does not use a physical mechanism such as a vibration table, the digital image can be vibrated at high speed. Therefore, in the case where the moving object system is moving at high speed, when the edge information of all objects in the digital image must be generated in real time, the present invention can achieve the required processing speed. According to the invention in the seventh scope of the patent application, Each image of the digital image is input in parallel to generate coarse edge information of each pixel of the digital image, and then the coarse edge information is output in parallel. The generated rough edge information is not necessarily correct because it is affected by the object's moving direction, moving speed, and color information difference between the object and the background, but it is not necessary to be special because of the noisy environment, such as outside the house with sunlight. The digital image is modified to generate rough edge information of a moving object of any shape. Therefore, the present invention can be used without being limited by the use environment. In addition, the present invention can treat infrared rays, ultraviolet rays, and even radiation rays in the same manner as the wavelengths of visible light and three primary colors, and the effect of noise can be reduced as the frequency band is increased. In this case, since the present invention does not need to change the circuit of the data processing device other than the memory capacity, the number of frequency bands can be easily increased. Therefore, the present invention can provide a contour, a position, and a size of a moving object at a high speed and inexpensively to a robot who can move not only inside the house but also outdoors. In addition, the present invention can also operate effectively for pre-processing of a visual recognition device that does not limit an object. According to the invention in the eighth scope of the patent application, the rough edge information and digital images can be input in parallel, and the digital images can be used from the rough edge information. 124 Jade paper scale is applicable to the Chinese National Standard (CNS) A4 specification (21〇 χ 297 mm) ^ -II n ϋ · ϋ · ϋ nn ϋ · n I 1 n ϋ i.— I -01 1 n I (Please read the notes on the back before filling this page) by 540010 Ministry of Economic Affairs Printed by the Intellectual Property Bureau employee consumer cooperative A7 B7_ 5. Description of the invention (Λ)) After generating and forming edge information, output it in parallel to form edge information. Since the present invention can form rough edge information generated by an arbitrary method into clearer and clearer edge information, it is possible to easily reduce the load required by a mechanism that generates rough edge information. In addition, since the present invention is less affected by rough edge information, the rough edge information generated by a specific field of a low-resolution low-resolution digital image can be set as a clearer and clearer edge information. Therefore, without increasing the amount of hardware and calculation, the present invention can generate high-precision edge information for objects in low-magnification digital images. With the present invention, an object recognition device using a wide-angle camera and a high-definition camera in the past can be used to implement object recognition with one camera. According to the invention in the ninth scope of the patent application, the edge information can be input in parallel, and the position and size of the object displayed by the edge information can be detected, and then the position and size of the object can be output in parallel as repeated information. According to the present invention, the quality of the position or size of a plurality of objects in an image detected by a large amount of hardware and calculations is required to be equal to or higher than the quality, and the edge information of the plurality of objects can be obtained from the edge information of the plurality of objects. Detected position and size. The present invention can also be used in the pre-processing of 'detecting the position and size of a plurality of objects in an image from a frame image of a moving image captured by a camcorder or a still image captured by a digital camera, Realize high-speed and low-cost pattern recognition algorithms for still images and moving images. In addition, according to the invention in the ninth scope of the patent application, it is possible to input the pixels of the object area image in parallel and detect the position and size of the object displayed by each pixel of the object area image for detection. Duplicate information is output side by side. The invention can apply the Chinese National Standard (CNS) A4 specification (21G X 297 public love) with a large amount of hardware and 125 I paper size " ------- Order --------- (Please read the notes on the back before filling out this page) 540010 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs Α7 Β7 V. Description of Invention (/ γ ) / The amount of position and size of the plurality of objects in the image detected by the calculation are of the same quality or above, and the position and size are detected from the edge information of the plurality of objects at one time. Especially when detecting the position and size of objects with round parts like cells or sperm, or objects that can be regarded as circular like human faces, it is not necessary to generate edge information again from these object fields, so it can be high speed and cheap Realize the pattern recognition algorithm of still images and dynamic images. According to the 10th invention of the scope of patent application, each pixel of the object area image and each pixel of the data image can be input in parallel to After the object area is normalized, each pixel of the normalized image is output in parallel. According to the present invention, the quality of the normalized image with multiple gaps generated by a large amount of hardware and calculation can be used, and the quality is the same or higher than that of the object area using the object area image. The object area of the occluded digital image is matched with the size of the digital image to compensate for the gap, and normalization is performed at the same time. It can also be used in the normalized pre-processing of the frame image of a moving image captured with a camcorder or a specific object cut out from a still image captured with a digital camera, and it can achieve stillness at high speed and low cost. Image recognition algorithms for images and moving images. According to the 11th invention in the scope of patent application, the pixels of several template images can be input in parallel, the pixels of the normalized image can be input in parallel, and after the template image and the graphics are matched, the pixels of the matching result image can be output in parallel. . If the reproducibility of the normalized image of the same object in different positions and sizes is high, the present invention can select a number of template images similar to the normalized image that are processed only in the vicinity. Therefore, since the present invention can apply the least square error to the 126 paper size to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) ----- f ------ install ------- -Order --------- ^ 9. (Please read the notes on the back before filling out this page) 540010 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7 V. Description of Invention (Λ ^) 〆 中The large-area processing of the sexual network is suppressed to a minimum, so it is possible to realize a high-speed and low-cost pattern recognition algorithm for still images and moving images. According to the invention in claim 12 of the scope of patent application, edge information can be formed in parallel input, and a non-linear oscillator is used to separate the object area from the background area, and then output each pixel of the object area image in parallel. Ben invented that for digital images, there is no need to perform any pre-processing other than the generation of edge information in advance, and it is not affected by the shape, position, and direction of the objects in the digital image. Moreover, even if the object fields in the digital image overlap, the object field and the background field can be separated. In addition, the present invention makes it easy to assemble hardware using digital technology, and also enables high-speed processing suitable for instant image processing. According to the inventions of claims 13 and 14, the visual device can be operated at high speed. Especially because of the parallelism, the processing time does not change even if the image size becomes larger. Therefore, visual devices can be used for applications that require immediateness. In addition, as long as the program is changed, the same chip can be used to implement a plurality of mechanisms in the range of patent application scope No. 6 to patent application scope No. 12, so the present invention can manufacture a vision device inexpensively. Furthermore, since the same mechanism can be used to implement a plurality of mechanisms from the 6th patent application to the 12th patent application, the wafer can be easily assembled in a mobile camera, thereby increasing convenience. Finally, since the present invention all uses pixel units to input and output signals, it is possible to suppress the amount of wiring by overlapping multiple wafers. Therefore, the present invention can easily improve the processability by using three-dimensional VLSI and other technological innovations. 127 The paper size is applicable to the Chinese National Standard (CNS) A4 specification (210 X 297 mm) (Please read the precautions on the back before filling this page. ) Pretend!
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP14563899 | 1999-05-25 | ||
| JP20973899 | 1999-07-23 | ||
| JP25099099 | 1999-09-06 | ||
| JP25098699 | 1999-09-06 | ||
| JP25363499 | 1999-09-07 |
| Publication Number | Publication Date |
|---|---|
| TW540010Btrue TW540010B (en) | 2003-07-01 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| TW89103654ATW540010B (en) | 1999-05-25 | 2000-03-02 | Vision device |
| Country | Link |
|---|---|
| TW (1) | TW540010B (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI772324B (en)* | 2016-09-12 | 2022-08-01 | 日商廣津生物化學股份有限公司 | A method for evaluating the chemotactic behavior of odorants based on the sense of smell of nematodes, and a petri dish and behavior evaluation system used for the evaluation method |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI772324B (en)* | 2016-09-12 | 2022-08-01 | 日商廣津生物化學股份有限公司 | A method for evaluating the chemotactic behavior of odorants based on the sense of smell of nematodes, and a petri dish and behavior evaluation system used for the evaluation method |
| Publication | Publication Date | Title |
|---|---|---|
| JP4324327B2 (en) | Visual equipment | |
| WO2022165809A1 (en) | Method and apparatus for training deep learning model | |
| US20180232583A1 (en) | Systems for generating parking maps and methods thereof | |
| Ishii et al. | Color-histogram-based tracking at 2000 fps | |
| Palaniappan et al. | Moving object detection for vehicle tracking in wide area motion imagery using 4d filtering | |
| Dellaert et al. | Jacobian images of super-resolved texture maps for model-based motion estimation and tracking | |
| CN112802049B (en) | A method and system for constructing a household object detection data set | |
| CN105488777A (en) | System and method for generating panoramic picture in real time based on moving foreground | |
| Tang et al. | FPGA‐Based Real‐Time Moving Target Detection System for Unmanned Aerial Vehicle Application | |
| CN113724527A (en) | Parking space management method | |
| Zhang et al. | Longitudinal-scanline-based arterial traffic video analytics with coordinate transformation assisted by 3D infrastructure data | |
| Xiao et al. | A real-time system for lane detection based on FPGA and DSP | |
| Mi et al. | Research on a Fast Human‐Detection Algorithm for Unmanned Surveillance Area in Bulk Ports | |
| JP2008158984A (en) | Image processing apparatus and area tracking program | |
| JP2011209896A (en) | Obstacle detecting apparatus, obstacle detecting method, and obstacle detecting program | |
| EP4481682A1 (en) | Spatial positioning method | |
| TW540010B (en) | Vision device | |
| CN100355284C (en) | visual device | |
| Zou et al. | Scene flow estimation by depth map upsampling and layer assignment for camera-LiDAR system | |
| Tian et al. | High confidence detection for moving target in aerial video | |
| JP2001148023A (en) | Visual device and visual method | |
| Stiller et al. | Fundamentals of machine vision | |
| Gandhi et al. | Motion analysis of omni-directional video streams for a mobile sentry | |
| JP2001148022A (en) | Visual device | |
| Shao et al. | Tracking algorithm using background-foreground motion models and multiple cues [surveillance video applications] |
| Date | Code | Title | Description |
|---|---|---|---|
| GD4A | Issue of patent certificate for granted invention patent | ||
| MM4A | Annulment or lapse of patent due to non-payment of fees |