CN100514359C - DSP based multiple channel mechanical digital display digital gas meter automatic detection device - Google Patents

Abstract

The invention discloses an automatic detecting device of the multiple-channel mechanical digital display gas meter based on DSP, which comprises the following parts: a cam for acquiring the image of the gas meter, a DSP processor for proceeding the digital identification of the cam image, wherein the DSP processor connected with the cam identifies the apparatus image acquired by the cam to the meter digit of the apparatus. The invention comprises the following steps: acquiring the digital display apparatus image by the image collecting system; positioning the meter interval with gamma component of YUV vector space; positioning the character position as BP nerval net input signal after divided by the character in the interval accurately and identifying the position of the decimal point (the BP nerval net applies the quick BP recognition algorithm of global self-adaptive belt momentum term). The invention provides an automatic detecting device of the multiple-channel mechanical digital display gas meter with higher identification ratio, quicker identification speed, and better convergence and better application based on DSP.

Description

Translated fromChinese

基于DSP的多通道机械数显数字煤气表自动检测装置Multi-channel mechanical digital display digital gas meter automatic detection device based on DSP

(一)技术领域(1) Technical field

本发明属于计算机视觉传感器技术、图像识别技术、计算机控制技术在自动化检测的应用，尤其是一种机械数显数字煤气表自动检测装置。The invention belongs to the application of computer vision sensor technology, image recognition technology and computer control technology in automatic detection, in particular to a mechanical digital display digital gas meter automatic detection device.

(二)背景技术(2) Background technology

现有的煤气表的用量，往往由抄表人员上门读取，目前市场上并不存在煤气表数字的自动检测装置。The consumption of the existing gas meter is often read by the meter reader at home, and there is no automatic detection device for the gas meter number on the market at present.

目前市场上其他的字符识别系统基本采用了“摄像头-采集卡-计算机”结构，实际是识别软件和通用图象设备的结合。这种系统结构复杂，成本高昂，安装不便，维护困难；软件的运行需要操作系统的支持，降低了系统的稳定性。At present, other character recognition systems on the market basically adopt the "camera-capture card-computer" structure, which is actually a combination of recognition software and general-purpose image equipment. This kind of system has complicated structure, high cost, inconvenient installation and difficult maintenance; the operation of the software needs the support of the operating system, which reduces the stability of the system.

(三)发明内容(3) Contents of the invention

为了克服已有的煤气表数字字符的识别装置的识别准确率低、识别速度慢、收敛性差、实时性差的不足，本发明提供一种识别准确率高、识别速度快、收敛性好、实时性好的以信号处理芯片(DSP)为平台的多通道机械数显数字煤气表自动检测装置。In order to overcome the shortcomings of the existing gas meter digital character recognition devices, such as low recognition accuracy, slow recognition speed, poor convergence and poor real-time performance, the present invention provides a high recognition accuracy, fast recognition speed, good convergence and real-time performance. A good multi-channel mechanical digital display digital gas meter automatic detection device based on a signal processing chip (DSP).

本发明解决其技术问题所采用的技术方案是：The technical solution adopted by the present invention to solve its technical problems is:

一种基于DSP的多通道机械数显数字煤气表自动检测装置，包括用于获取煤气表的图像的摄像头、对摄像头图像进行数字识别的DSP处理器，所述摄像头连接DSP处理器，所述的DSP处理器包括：图像采集模块，用于采集摄像头的数据，数显仪表的图像格式为彩色模拟视频信号；所述的DSP处理器还包括：仪表表头数字区定位模块，用于将获取的原始模拟视频信号经过解码后，取出图像中的色调V分量，利用阈值分割图象，确定仪表表头数字区；数字字符去噪模块，用于将得到的仪表表头数字区去噪；数字字符分割模块，用于将仪表表头数字区转化成像素直方图，并分割成数字字符；小数点定位模块，用于将分割后字符的右下角包含小数点区域的白色像素进行累加，累加的和除以小数点区域内的像素总数，如比例大于50％，判断存在小数点；字符特征量抽取模块，用于将分割出来的噪声区域去掉，字符的宽度为WIDTH，高度为HEIGHT，如分割后字符图像的WIDTH/HEIGHT<1.2，删除该字符，并对分割后的各个字符图像进行编码，编码的规则为：用2*1的网格将数字目标区域进行相交，用得到的交点进行数字判断，提取各个数字图像的特征编码；数字字符识别模块，用于将提取的特征编码送到BP神经网络训练或识别，得到对应该特征编码的数字。A DSP-based multi-channel mechanical digital display digital gas meter automatic detection device, including a camera for obtaining the image of the gas meter, a DSP processor for digitally recognizing the camera image, the camera is connected to the DSP processor, and the The DSP processor includes: an image acquisition module, which is used to collect the data of the camera, and the image format of the digital display instrument is a color analog video signal; the DSP processor also includes: an instrument gauge digital area positioning module, which is used to obtain the After the original analog video signal is decoded, the tone V component in the image is taken out, and the image is segmented by a threshold to determine the digital area of the meter head; the digital character denoising module is used to denoise the obtained digital area of the meter head; the digital character The segmentation module is used to convert the digital area of the instrument meter into a pixel histogram and divide it into digital characters; the decimal point positioning module is used to accumulate the white pixels containing the decimal point area in the lower right corner of the divided character, and divide the accumulated sum by The total number of pixels in the decimal point area, if the ratio is greater than 50%, it is judged that there is a decimal point; the character feature quantity extraction module is used to remove the segmented noise area, the width of the character is WIDTH, and the height is HEIGHT, such as the WIDTH of the character image after segmentation /HEIGHT<1.2, delete the character, and encode each character image after segmentation. The encoding rule is: intersect the digital target area with a 2*1 grid, use the obtained intersection point to judge the number, and extract each number The feature code of the image; the digital character recognition module is used to send the extracted feature code to the BP neural network for training or recognition, and obtain the number corresponding to the feature code.

进一步，在所述的字符特征量抽取模块中，如WIDTH/HEIGHT>3，且将白色像素进行横向累加，把累加后的白色像素的比例大于0.3，判断为数字1；设图像的高度为H，宽度为W，令区间(0，H/4)为h1，区间(H/4，3H/4)为h2，区间(3H/4，H)为h3，区间(0，W/2)为w1，区间(W/2，W)为w2，编码如下：Further, in the character feature extraction module, if WIDTH/HEIGHT>3, and the white pixels are horizontally accumulated, the ratio of the accumulated white pixels is greater than 0.3, and it is judged as anumber 1; the height of the image is H , the width is W, let the interval (0, H/4) be h1, the interval (H/4, 3H/4) be h2, the interval (3H/4, H) be h3, and the interval (0, W/2) be w1, the interval (W/2, W) is w2, and the encoding is as follows:

  (W/2，h1) (W/2，h2) (W/2，h3) (w1，H/4) (w2，H/4) (w1，3H/4) (w2，3H/4)(W/2, h1) (W/2, h2) (W/2, h3) (w1, H/4) (w2, H/4) (w1, 3H/4) (w2, 3H/4)

0：  1          0         1         1         1         1          10: 1 0 1 1 1 1 1 1 1

2：  1          1         1         0         1         1          02: 1 1 1 1 0 1 1 1 0

3：  1          1         1         0         0         1          13: 1 1 1 1 0 0 0 1 1 1

4：  0          1         0         1         0         1          14: 0 1 0 1 1 0 1 1 1

5：  1          1         1         1         0         0          15: 1 1 1 1 1 0 0 0 1

6：  1          1         1         1         1         0          16: 1 1 1 1 1 1 1 0 1

7：  1          0         0         0         0         1          17: 1 0 0 0 0 0 0 1 1 1

8：  1          1         1         1         1         1          18: 1 1 1 1 1 1 1 1 1 1

9：  1          1         1         1         0         1          1。9: 1 1 1 1 1 1 0 0 1 1 .

再进一步，所述的仪表表头数字区定位模块包括：Still further, the positioning module of the digital area of the meter head includes:

粗定位单元，用于得到二值化图像后，将象素按行和列累加并将结果存在行数组和列数组中，然后对行数组进行上下包夹，得到行目标矩形；然后对列数组进行左右包夹，得到列目标矩形；The coarse positioning unit is used to accumulate the pixels by row and column after obtaining the binarized image and store the result in the row array and column array, and then double-enclose the row array to obtain the row target rectangle; then the column array Perform left and right double-teaming to obtain the column target rectangle;

精确定位单元，用于将象素行累加，计算出目标数目n，并将坐标存入坐标数组：The precise positioning unit is used to accumulate the pixel rows, calculate the target number n, and store the coordinates into the coordinate array:

(1)判断n是否为0，如果是，转到步骤(2.2.3)，如果否，判断每个横向目标区域是否存在其他数字目标，假如有，设为m_i，m_i表示第i个横向区域内的数字目标数，将坐标存在坐标数组；(1) Determine whether n is 0, if so, go to step (2.2.3), if not, determine whether there are other digital targets in each horizontal target area, if so, set it to m_i , m_i represents the i-th The number of digital targets in the horizontal area, save the coordinates in the coordinate array;

(2)、计算总数目N＝m₁+m₂+...+m_i+...+m_n，得到目标数组，分割为相应矩形区域分别为表头的坐标；(2) Calculate the total number N=m₁ +m₂ +...+m_i +...+m_n to obtain the target array, which is divided into corresponding rectangular areas as the coordinates of the header;

(3)、输入为表头区域图像，使用双线性插值法放大图像。(3) The input is the image of the table header area, and the image is enlarged by bilinear interpolation.

更进一步，所述的仪表表头数字区定位模块还包括：颜色选取单元，用于如果表头字符为红色，使用RED分量图进行颜色选取，如果表头字符为蓝色，使用BLUE分量进行颜色选取。Further, the meter meter header digital area positioning module also includes: a color selection unit, for if the header character is red, use the RED component map for color selection, if the meter header character is blue, use the BLUE component for color selection select.

在所述的仪表表头数字区定位模块中，从RGB向量空间到YUV向量空间的转换参见公式(1)：Y＝(R+16)*0.257+(G+128)*0.504+(B+128)*0.098In the digital area positioning module of the instrument meter head, the conversion from the RGB vector space to the YUV vector space is referred to in formula (1): Y=(R+16)*0.257+(G+128)*0.504+(B+ 128)*0.098

U＝(R+16)*(-0.148)+(G+128)*(-0.291)+(B+128)*0.439U＝(R+16)*(-0.148)+(G+128)*(-0.291)+(B+128)*0.439

V＝(R+16)*0.439+(G+128)*-0.368+(B+128)*-0.071        (1)；V＝(R+16)*0.439+(G+128)*-0.368+(B+128)*-0.071 (1);

其中V分量(val_V)代表色调，用阈值VAL将图像二值化，即：Among them, the V component (val_V) represents the hue, and the image is binarized with the threshold VAL, namely:

pic_V＝“1”when val_V>VALpic_V="1" when val_V>VAL

else pic_V“0”，else pic_V "0",

其中pic_V是得到的二值化结果图像；V分量的最大值和最小值分别是Vmax和Vmin，阈值VAL＝(Vmax+Vmin)/2，得到二值化图像。Wherein pic_V is the obtained binarization result image; the maximum and minimum values of the V component are Vmax and Vmin respectively, and the threshold VAL=(Vmax+Vmin)/2 to obtain a binarization image.

所述的DSP处理器还包括：图像实时显示模块，用于将采集的数据进行显示；图像数据文件存储模块，用于将采集的数据进行存储；数字字符实时显示模块，用于将识别后的数字进行显示。Described DSP processor also comprises: image real-time display module, is used for displaying the data collected; Image data file storage module, is used for storing the data collected; Digital character real-time display module, is used for identifying after The numbers are displayed.

本发明的有益效果主要表现在：1、识别准确率高；2、识别速度快；3、收敛性好；4、具有嵌入性；5、能对多台数显仪表动态变化的数字字符和小数点的位置进行实时与并行识别和定位，且仪表的位置可变化；6、小数点的定位准确。The beneficial effects of the present invention are mainly manifested in: 1. High recognition accuracy; 2. Fast recognition speed; 3. Good convergence; 4. Embedded; Real-time and parallel identification and positioning of the position, and the position of the instrument can be changed; 6. The positioning of the decimal point is accurate.

(四)附图说明(4) Description of drawings

图1是基于DSP的数显仪表动态显示字符识别装置的原理框图。Figure 1 is a functional block diagram of a DSP-based digital display instrument dynamic display character recognition device.

图2是基于DSP的数显仪表动态显示字符识别装置的识别流程图。Fig. 2 is a recognition flow chart of a DSP-based digital display instrument dynamic display character recognition device.

图3是识别后的图像示意图。Fig. 3 is a schematic diagram of the recognized image.

(五)具体实施方式(5) Specific implementation methods

下面结合附图对本发明作进一步描述。The present invention will be further described below in conjunction with the accompanying drawings.

参照图1—图3，一种基于DSP的多通道机械数显数字煤气表自动检测装置，包括用于获取煤气表的图像的摄像头1、对摄像头图像进行数字识别的DSP处理器2，摄像头采用CCD，所述摄像头1连接DSP处理器2，所述的DSP处理器2包括：图像采集模块3，用于采集摄像头的数据，数显仪表的图像格式为彩色模拟视频信号；所述的DSP处理器还包括：仪表表头数字区定位模块4，用于将获取的原始模拟视频信号经过解码后，取出图像中的色调V分量，利用阈值分割图象，确定仪表表头数字区；数字字符去噪模块5，用于将得到的仪表表头数字区去噪；数字字符分割模块6，用于将仪表表头数字区转化成像素直方图，并分割成数字字符；小数点定位模块7，用于将分割后字符的右下角包含小数点区域的白色像素进行累加，累加的和除以小数点区域内的像素总数，如比例大于50％，判断存在小数点；字符特征量抽取模块8，用于将分割出来的噪声区域去掉，字符的宽度为WIDTH，高度为HEIGHT，如分割后字符图像的WIDTH/HEIGHT<1.2，删除该字符，并对分割后的各个字符图像进行编码，编码的规则为：用2*1的网格将数字目标区域进行相交，用得到的交点进行数字判断，提取各个数字图像的特征编码；数字字符识别模块9，用于将提取的特征编码送到BP神经网络训练或识别，得到对应该特征编码的数字。With reference to Fig. 1-Fig. 3, a kind of multi-channel mechanical digital display digital gas meter automatic detection device based on DSP, comprises thecamera 1 that is used to obtain the image of gas meter, theDSP processor 2 that digital recognition is carried out to camera image, camera adopts CCD, describedcamera 1 connectsDSP processor 2, and describedDSP processor 2 comprises:image acquisition module 3, is used to gather the data of camera, and the image format of digital display instrument is a color analog video signal; Described DSP processes The device also includes: an instrument gauge digitalarea positioning module 4, which is used to decode the obtained original analog video signal, take out the tone V component in the image, and utilize the threshold value to segment the image to determine the instrument gauge digital area; digital characters are removedNoise module 5, is used for the denoising of the digital area of the instrument gauge that obtains; Digitalcharacter segmentation module 6, is used for converting the digital area of the instrument gauge into a pixel histogram, and is divided into digital characters; Decimalpoint positioning module 7, for Accumulate the white pixels in the lower right corner of the character after segmentation that includes the decimal point area, and divide the accumulated sum by the total number of pixels in the decimal point area. If the ratio is greater than 50%, it is judged that there is a decimal point; the character featurequantity extraction module 8 is used to separate The noise area is removed, the width of the character is WIDTH, and the height is HEIGHT. If the character image after segmentation is WIDTH/HEIGHT<1.2, delete the character and encode each character image after segmentation. The encoding rule is: Use 2* The grid of 1 intersects the digital target area, uses the obtained intersection point to carry out digital judgment, and extracts the feature codes of each digital image; the digitalcharacter recognition module 9 is used to send the extracted feature codes to the BP neural network for training or identification, and obtains The number corresponding to this feature encoding.

所述的DSP处理器还包括：图像实时显示模块10，用于将采集的数据进行显示；图像数据文件存储模块11，用于将采集的数据进行存储；数字字符实时显示模块12，用于将识别后的数字进行显示。所述图像实时显示模块10和数字字符实时显示模块12连接显示器。Described DSP processor also comprises: image real-time display module 10, is used for displaying the data collected; Image datafile storage module 11, is used for storing the data collected; Digital character real-time display module 12, is used for displaying The recognized numbers are displayed. The image real-time display module 10 and the digital character real-time display module 12 are connected to a display.

在DSP处理器中，PRD_Ticks是系统内部时钟模块使用的标准时钟，每格为1ms，tskLoopback就是识别函数，从图中可以看出，整个识别过程大概需要4ms，再加上前面计算所得的传输时间，整个DSP程序的耗时为5.7ms，这远远满足实时要求。表1就是系统各个模块的处理时间：In the DSP processor, PRD_Ticks is the standard clock used by the internal clock module of the system, each division is 1ms, tskLoopback is the recognition function, as can be seen from the figure, the entire recognition process takes about 4ms, plus the transmission time calculated earlier , the time-consuming of the entire DSP program is 5.7ms, which is far from meeting the real-time requirements. Table 1 is the processing time of each module of the system:

 功能处理时间(ms)    捕获图像1.36识别过程4显示图像0.34 Function processing time (ms) capture image 1.36identification process 4 display image 0.34

表1Table 1

本实施例的字符识别方法包括以下步骤：The character recognition method of the present embodiment comprises the following steps:

(1)、通过摄像头获取煤气表的图像，格式为彩色模拟视频信号；(1), the image of the gas meter is obtained through the camera, and the format is a color analog video signal;

(2)、仪表表头数字区定位：(2) Positioning of the digital area of the meter head:

粗定位输入的是原始图像，经过解码后获得YUV向量空间，定位到目标区域。The coarse positioning input is the original image, and after decoding, the YUV vector space is obtained and positioned to the target area.

其中V分量(val_V)代表色调，用阈值VAL将图像二值化，即Among them, the V component (val_V) represents the hue, and the image is binarized with the threshold VAL, that is

pic_V＝“1”when val_V>VALpic_V="1" when val_V>VAL

elsepic_V“0”，else pic_V "0",

其中pic_V是得到的二值化结果图像。Where pic_V is the obtained binarized result image.

其中V分量的最大值和最小值分别是Vmax和Vmin，阈值VAL＝(Vmax+Vmin)/2；Wherein the maximum value and the minimum value of the V component are respectively Vmax and Vmin, and the threshold value VAL=(Vmax+Vmin)/2;

(2.1)、粗定位的算法：(2.1), coarse positioning algorithm:

将象素按行和列累加并将结果存在VVHeightD和VVWidthD这两个数组里面，然后对VVHeightD进行上下包夹，得到目标矩形rect.top和rect.bottom，然后对VVWidthD进行左右包夹，得到rect.left和rect.right.原始图象粗定位结果为：rect.left，rect.right，rect.top，rect.bottom。Accumulate the pixels by row and column and store the result in the two arrays VVHeightD and VVWidthD, then wrap VVHeightD up and down to get the target rectangle rect.top and rect.bottom, and then wrap VVWidthD left and right to get rect .left and rect.right. The rough positioning results of the original image are: rect.left, rect.right, rect.top, rect.bottom.

(2.2)、精确定位的算法：(2.2), precise positioning algorithm:

输入为pic_V中粗定位的目标区域，将象素行累加，计算出目标数目n，并将坐标存入坐标数组pic_rt[n]The input is the rough positioning target area in pic_V, the pixel rows are accumulated, the target number n is calculated, and the coordinates are stored in the coordinate array pic_rt[n]

(2.2.1)、判断n是否为0，如果是转到步骤(2.2.3)，如果否，判断每个横向目标区域是否存在其他数字目标，假如有，设为m₁，m₁表示第i个横向区域内的数字目标数，并将坐标存在坐标数组pic_rt[n][m]里，n＝n-1，转到步骤2).(2.2.1), judge whether n is 0, if it is to go to step (2.2.3), if not, judge whether there are other digital targets in each horizontal target area, if there is, set it as m₁ , m₁ means the first The number of digital targets in the i horizontal area, and store the coordinates in the coordinate array pic_rt[n][m], n=n-1, go to step 2).

(2.2.2)、计算总数目N＝m₁+m₂+...+m₁+...+m_n，并且得到目标数组pic_rt[n][m]，分割为相应矩形区域，分别为表头的坐标。(2.2.2), calculate the total number N=m₁ +m₂ +...+m₁ +...+m_n , and obtain the target array pic_rt[n][m], which is divided into corresponding rectangular areas, respectively is the coordinates of the header.

(2.2.3)、表头区域放大：输入为表头区域图像，图像放大使用双线性插值法；每两个象素之间插了3个值，即相应放大9倍，因为在原始图像中，一个字符的宽和高只有10个象素左右，影响小数点的判别，而且双线性插值法又称一阶插值法，不会产生锯齿现象。(2.2.3) Enlargement of the header area: the input is the image of the header area, and the image is enlarged using the bilinear interpolation method; 3 values are inserted between every two pixels, that is, the corresponding magnification is 9 times, because in the original image In , the width and height of a character are only about 10 pixels, which affects the judgment of the decimal point, and the bilinear interpolation method is also called the first-order interpolation method, which will not produce aliasing.

(2.3)、颜色选取：表头字符一般为红色，后续处理中使用RED分量图进行处理，相应的如果字符区域是蓝色的，将使用BLUE分量进行处理。(2.3), color selection: the header characters are generally red, and the RED component map is used for subsequent processing. Correspondingly, if the character area is blue, the BLUE component will be used for processing.

(3)、去噪过程：(3), denoising process:

由于系统受光线和背景的影响比较严重，因此利用目标数字区域上方的一个4*4像素区域内红色分量的均值为阈值，消除外界的干扰。Since the system is seriously affected by light and background, the average value of the red component in a 4*4 pixel area above the target digital area is used as the threshold to eliminate external interference.

去噪算法：选定目标点，判断周围8个象素中的白色象素的数目，如果白色象素的数目超过4个，则目标点为白色，即是”1”.。Denoising algorithm: Select the target point and judge the number of white pixels in the surrounding 8 pixels. If the number of white pixels exceeds 4, the target point is white, which is "1".

(4)、数字字符分割过程：(4), digital character segmentation process:

首先将图8进行列累加，把结果存在数组DDWidth里；First, accumulate the columns of Figure 8, and store the result in the array DDWidth;

(5)、小数点定位过程：(5), decimal point positioning process:

将分割后的字符的右下角包含小数点区域(图中红色区域)的白色象素进行累加，除以区域内的象素总数，如果比例大于50％，判断为存在小数点。The lower right corner of the character after the segmentation contains the white pixels of the decimal point area (red area in the figure) are accumulated, divided by the total number of pixels in the area, if the ratio is greater than 50%, it is judged that there is a decimal point.

(6)、字符特征量抽取过程：(6), character feature quantity extraction process:

首先把分割出来的一些噪声区域去掉，归为以下两类，分割后图像的WIDTH/HEIGHT>1.2，由于数字一般HEIGHT比WIDTH大，所以我把这种情况去掉.First remove some of the segmented noise areas and classify them into the following two categories. The WIDTH/HEIGHT of the segmented image is > 1.2. Since the digital HEIGHT is generally larger than the WIDTH, I will remove this situation.

(7)、对分割后的各个字符图像进行编码，编码的规则为：用2*1的网格将数字目标区域进行相交，用得到的交点进行数字判断，提取各个数字图像的特征编码；(7), each character image after segmentation is encoded, and the rule of encoding is: intersect the digital target area with the grid of 2*1, carry out digital judgment with the intersection point that obtains, extract the feature code of each digital image;

如HEIGHT/WIDTH>3，这种情况只有在出现数字是1的情况，但是也会出现噪声，比如上面图片中介于第5的数字4和第6个数字0之间的那个白点，所以我将白色象素进行横向累加，把累加后白色象素的比例<0.3的当成噪声点，如果比例>0.3则判断是1；Such as HEIGHT/WIDTH>3, this situation is only when the number is 1, but there will also be noise, such as the white point between the5th number 4 and the 6th number 0 in the picture above, so I The white pixels are accumulated horizontally, and the ratio of the accumulated white pixels <0.3 is regarded as a noise point, and if the ratio>0.3, it is judged to be 1;

提取特征算法为：用2*1的网格将数字目标区域进行相交，用得到的交点进行数字判断：The feature extraction algorithm is: use a 2*1 grid to intersect the digital target area, and use the obtained intersection point for digital judgment:

设图像高为H，宽为W；令区间(0，H/4)为h1，区间(H/4，3H/4)为h2，区间(3H/4，H)为h3，区间(0，W/2)为w1，区间(W/2，W)为w2；则编码如下：Let the image height be H and the width be W; let the interval (0, H/4) be h1, the interval (H/4, 3H/4) be h2, the interval (3H/4, H) be h3, and the interval (0, W/2) is w1, and the interval (W/2, W) is w2; the encoding is as follows:

   (W/2，h1) (W/2，h2) (W/2，h3) (w1，H/4) (w2，H/4) (w1，3H/4) (w2，3H/4)(W/2, h1) (W/2, h2) (W/2, h3) (w1, H/4) (w2, H/4) (w1, 3H/4) (w2, 3H/4)

0：    1         0         1         1         1         1          10: 1 0 0 1 1 1 1 1 1 1

2：    1         1         1         0         1         1          02: 1 1 1 1 0 1 1 1 0

3：    1         1         1         0         0         1          13: 1 1 1 1 0 0 0 1 1 1

4：    0      1      0      1     0     1      14: 0 1 0 0 1 0 0 1 1

5：    1      1      1      1     0     0      15: 1 1 1 1 1 0 0 0 1

6：    1      1      1      1     1     0      16: 1 1 1 1 1 1 1 0 1

7：    10            0      0     0     1      17: 10 0 0 0 0 1 1

8：    1      1      1      1     1     1      18: 1 1 1 1 1 1 1 1 1

9：    1      1      1      1     0     1      19: 1 1 1 1 1 0 1 1

(8)、将提取的特征编码送到BP神经网络训练或识别，得到对应该特征编码的数字。(8), the extracted feature code is sent to the BP neural network for training or recognition, and the number corresponding to the feature code is obtained.

Claims (6)

1, a kind of multiple channel mechanical digital display digital gas meter automatic detection device based on DSP, comprise the camera of the image that is used to obtain gas meter, camera image carried out the dsp processor of digit recognition, described camera connects dsp processor, and described dsp processor comprises:

Image capture module is used for the data of acquisition camera, and the picture format of digital displaying meter is colored analog video signal;

It is characterized in that: described dsp processor also comprises:

Dial of instrument digital block locating module, the original analog vision signal that is used for obtaining is taken out the tone V component in the image through after decoding, and utilizes the Threshold Segmentation image, determines the Dial of instrument digital block;

Numerical character denoising module is used for the Dial of instrument digital block denoising that will obtain;

Numerical character is cut apart module, is used for the Dial of instrument digital block is changed into the pixel histogram, and is divided into numerical character;

The radix point locating module, be used for will cut apart the lower right corner of the back character white pixel that comprises the radix point zone add up, add up and divided by the sum of all pixels in the radix point zone,, judge to have radix point greater than 50% as ratio; Character feature amount abstraction module, be used for the noise region that splits is removed, the width of character is WIDTH, be HEIGHT highly, as cut apart WIDTH/HEIGHT＜1.2 of back character picture, delete this character, and each character picture after cutting apart encoded, the rule of coding is: the grid with 2*1 intersects the digital object zone, carries out numeral with the intersection point that obtains and judges, extracts the feature coding of each digital picture;

The Number character recognition module is used for the feature coding that extracts is delivered to BP neural metwork training or identification, obtains numeral that should feature coding.

2, the multiple channel mechanical digital display digital gas meter automatic detection device based on DSP as claimed in claim 1, it is characterized in that: in described character feature amount abstraction module, as WIDTH/HEIGHT〉3, and white pixel is laterally added up, the ratio of the white pixel after adding up greater than 0.3, is judged as numeral 1;

If the height of image is H, width is W, the order interval (0, H/4) be h1, interval (H/4 3H/4) is h2, interval (3H/4 H) is h3, interval (0, W/2) be w1, interval (W/2 W) is w2, and it is as follows to encode:

(W/2，h1) W/2，h2) (W/2，h3) (w1，H/4) (w2，H/4) (w1，3H/4) (w2，3H/4)

0： 1 0 1 1 1 1 1

2： 1 1 1 0 1 1 0

3： 1 1 1 0 0 1 1

4： 0 1 0 1 0 1 1

5： 1 1 1 1 0 0 1

6： 1 1 1 1 1 0 1

7： 1 0 0 0 0 1 1

8： 1 1 1 1 1 1 1

9： 1 1 1 1 0 1 1。

3, the multiple channel mechanical digital display digital gas meter automatic detection device based on DSP as claimed in claim 1 or 2 is characterized in that: described Dial of instrument digital block locating module comprises:

The coarse positioning unit after being used to obtain binary image, adds up pixel and the result is existed in line number group and the columns group by row and column, then the line number group is carried out double team up and down, obtains the row target rectangle; Then the columns group is carried out left and right sides double team, obtain the row target rectangle;

Accurately positioning unit is used for pixel rows is added up, and calculates target numbers n, and deposits coordinate in the coordinate array:

(1) judge whether n is 0, if, forward (3) to, if not, judge whether each horizontal target area exists other digital objects, if having, is made as mi, mi represents the digital object number in i the transverse area, and there is the coordinate array in coordinate;

(2), the order N=m that totalizes₁+ m₂+ ...+m_i+ ...+m_n, obtain the target array, be divided into the coordinate that the respective rectangular zone is respectively gauge outfit;

(3), be input as gauge outfit area image, use bilinear interpolation enlarged image.

4, the multiple channel mechanical digital display digital gas meter automatic detection device based on DSP as claimed in claim 3,

It is characterized in that: described Dial of instrument digital block locating module also comprises:

Color is chosen the unit, if be used for the gauge outfit character for red, use RED to divide spirogram to carry out color and chooses, if the gauge outfit character is blue, uses the BLUE component to carry out color and chooses.

5, the multiple channel mechanical digital display digital gas meter automatic detection device based on DSP as claimed in claim 4, it is characterized in that: in the locating module of described Dial of instrument digital block, the conversion from the RGB vector space to the YUV vector space is referring to formula (1): Y=(R+16) * 0.257+ (G+128) * 0.504+ (B+128) * 0.098U=(R+16) * (0.148)+(G+128) * (0.291)+(B+128) * 0.439

V＝(R+16)*0.439+(G+128)*-0.368+(B+128)*-0.071 (1)；

Wherein V component val_V represents tone, uses threshold value VAL with image binaryzation, that is:

pic_V＝“1”whenVal_V>VAL

else pic_V＝“0”，

Wherein pic_V is the binaryzation result images that obtains; The maximal value of V component and minimum value are respectively Vmax and Vmin, and threshold value VAL=(Vmax+Vmin)/2 obtains binary image.

6, the multiple channel mechanical digital display digital gas meter automatic detection device based on DSP as claimed in claim 3,

It is characterized in that: described dsp processor also comprises:

The real-time display module of image, the data that are used for gathering show;

The image data file memory module is used for the data of gathering are stored;

The real-time display module of numerical character is used for the numeral after the identification is shown.

Images