Movatterモバイル変換

[0]ホーム
URL:

CN102542289A - Pedestrian volume statistical method based on plurality of Gaussian counting models - Google Patents

Pedestrian volume statistical method based on plurality of Gaussian counting modelsDownload PDF

Info

Publication number
CN102542289A
CN102542289ACN2011104233492ACN201110423349ACN102542289ACN 102542289 ACN102542289 ACN 102542289ACN 2011104233492 ACN2011104233492 ACN 2011104233492ACN 201110423349 ACN201110423349 ACN 201110423349ACN 102542289 ACN102542289 ACN 102542289A
Authority
CN
China
Prior art keywords
gaussian
counting
moving target
image
people
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN2011104233492A
Other languages
Chinese (zh)
Other versions
CN102542289B (en
Inventor
高陈强
余迪虎
李璐星
李强
查力
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chongqing University of Post and Telecommunications
Original Assignee
Chongqing University of Post and Telecommunications
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Chongqing University of Post and TelecommunicationsfiledCriticalChongqing University of Post and Telecommunications
Priority to CN201110423349.2ApriorityCriticalpatent/CN102542289B/en
Publication of CN102542289ApublicationCriticalpatent/CN102542289A/en
Application grantedgrantedCritical
Publication of CN102542289BpublicationCriticalpatent/CN102542289B/en
Activelegal-statusCriticalCurrent
Anticipated expirationlegal-statusCritical

Links

Images

Landscapes

Abstract

The invention relates to intelligent video surveillance and image processing and analysis, discloses a pedestrian volume statistical method, and comprises establishing a plurality of Gaussian counting models by utilizing training video sequence image samples with people number marks and performing real-time pedestrian volume statistics on videos with unknown people numbers based on the plurality of Gaussian counting models. The pedestrian volume statistical method particularly comprises the steps of firstly extracting a prospect moving target according to moving target detection, extracting eigenvectors according to moving target area and characteristics including lengths and widths of an external rectangular frame, then establishing the plurality of Gaussian counting models based on an eigenvector set, and finally analyzing numbers of pedestrians contained in an unknown moving target area based on the plurality of Gaussian counting models to achieve pedestrian volume statistics. By establishing the plurality of Gaussian counting models, the pedestrian volume statistical method avoids difficulties caused by identification and tracking of singe pedestrian, can perform statistics of the numbers of the pedestrians contained in moving target areas in different detection areas well, improves statistical accuracy of the numbers of the pedestrians, and then improves accuracy of the pedestrian volume statistics.

Description

Translated fromChinese
一种基于多高斯计数模型的人流量统计方法A People Flow Statistics Method Based on Multi-Gaussian Counting Model

技术领域technical field

本发明涉及图像处理技术领域，尤其涉及一种人流量统计的方法及系统。The present invention relates to the technical field of image processing, in particular to a method and system for counting people flow.

背景技术Background technique

现有的视频监控中，大多数只是简单的实现视频传输，然后依靠人眼观察来实现场景监控以及计数等工作。这种人力监控方式存在大量的不足之处，如比较枯燥，监控人员很容易产生疲劳而导致工作失误，此外，随着人力成本提高，依靠人力监控计数这一手段将不再适合。In the existing video surveillance, most of them simply realize video transmission, and then rely on human eye observation to realize scene monitoring and counting. This human monitoring method has a lot of shortcomings, such as being boring, and the monitoring personnel are prone to fatigue and lead to work mistakes. In addition, with the increase of labor costs, relying on the method of manpower monitoring and counting will no longer be suitable.

目前基于计算机视觉的人流量统计系统所采用的方法可以分为三类：一是基于行人检测跟踪的方法；二是基于特征点轨迹聚类的方法；三是基于低层特征回归的方法。At present, the methods used in the computer vision-based people flow statistics system can be divided into three categories: one is the method based on pedestrian detection and tracking; the other is the method based on feature point trajectory clustering; the third is the method based on low-level feature regression.

基于行人检测跟踪的方法的核心在于多目标检测，这种方法是通过背景差分或者机器学习的方法得到前景区域，然后采用运动形态联合分割或者模板匹配的方法完成人流量统计的任务。通常情况下，该算法可以获得较高的检测精度。如申请号为201010114826.2的中国专利申请文件中所提到的方法，首先通过分类器对当前图像进行人头粗检测，然后对粗检测结果进行边缘特征细筛选处理，虽然通过上述方法能够有效提高人头的检测率，但是，当人群密度较高，发生遮挡等情况时，存在人头漏检，或者多检的情况，最终导致检测结果不够准确，而且该方法计算量较大，难以实时处理。The core of the method based on pedestrian detection and tracking is multi-target detection. This method obtains the foreground area through background difference or machine learning, and then uses joint segmentation of motion patterns or template matching to complete the task of counting people. Usually, this algorithm can obtain higher detection accuracy. For example, the method mentioned in the Chinese patent application document with the application number 201010114826.2 first uses a classifier to perform rough head detection on the current image, and then performs fine screening of edge features on the rough detection results. Although the above method can effectively improve the head However, when the crowd density is high and occlusions occur, there may be cases of missed detection or multiple detections, which will eventually lead to inaccurate detection results. Moreover, this method requires a large amount of calculation and is difficult to process in real time.

基于特征点轨迹聚类的方法首先通过跟踪一些特征点，对具有一致运动特性的特征点轨迹进行聚类分析达到人数统计的目的。该算法能够有效减少摄像机视角的影响。然而，特征点本身难以稳定可靠的跟踪，因此该算法统计精度比较低。The method based on feature point trajectory clustering first tracks some feature points, and clusters and analyzes the feature point trajectories with consistent motion characteristics to achieve the purpose of population counting. This algorithm can effectively reduce the impact of camera angle of view. However, the feature points themselves are difficult to track stably and reliably, so the statistical accuracy of the algorithm is relatively low.

基于低层特征回归的方法首先利用背景差分得到前景区域，然后计算前景区域中的特征如面积、边缘、纹理等，最后通过各种回归函数如线性、高斯过程回归、神经网络等建立特征与人流量的函数关系。该算法跳过了对于单个行人目标的检测跟踪过程，降低了计算复杂度，在一定程度上可以达到实时性要求。但是其通用性不够理想，而且统计精确度与前景像素提取的依赖关系较大，因此该方法难以获得精确的人数信息。The method based on low-level feature regression first uses background difference to obtain the foreground area, then calculates the features in the foreground area such as area, edge, texture, etc., and finally establishes features and traffic flow through various regression functions such as linear, Gaussian process regression, neural network, etc. functional relationship. The algorithm skips the detection and tracking process of a single pedestrian target, reduces the computational complexity, and can meet the real-time requirements to a certain extent. However, its versatility is not ideal, and the statistical accuracy is largely dependent on the extraction of foreground pixels, so it is difficult to obtain accurate information about the number of people in this method.

综上所述，现有技术中的人流量统计方法中难点主要在于如何对具有较大密度的流动人群进行较高精度的统计，且算法的复杂度不能过高，并满足实时应用需求。To sum up, the difficulty in the people flow counting method in the prior art is how to count the floating crowd with a relatively high density with high accuracy, and the complexity of the algorithm should not be too high, and meet the real-time application requirements.

发明内容Contents of the invention

本发明针对现有基于计算机视觉的人流量统计技术中存在的上述问题，提出一种基于多高斯计数模型的实时人流量统计方法，以解决现有人流量统计方案对较高密度流动人群统计不准确的问题。The present invention aims at the above-mentioned problems existing in the existing computer vision-based people flow statistics technology, and proposes a real-time people flow statistics method based on a multi-Gaussian counting model, so as to solve the inaccurate statistics of relatively high-density mobile crowds in the existing people flow statistics scheme The problem.

本发明解决上述技术问题采用如下技术方案：The present invention solves above-mentioned technical problem and adopts following technical scheme:

一种基于多高斯计数模型的人流量统计方法，包括：输入图像预处理、运动目标检测、运动目标特征向量提取和多高斯计数模型建立、运动目标跟踪、以及人流量统计等步骤。具体为：A method for counting people flow based on a multi-Gaussian counting model, comprising: input image preprocessing, moving object detection, moving object feature vector extraction, multi-Gaussian counting model establishment, moving object tracking, and people flow counting steps. Specifically:

在实际检测中，往往只对场景中某区域感兴趣，因此首先选取兴趣区域(region of interest , ROI)，后续的所有图像处理操作均在该兴趣区域内完成。将兴趣区域分割为多个面积大小相等的检测子区域，采用前景图像与背景图像作差分的运动目标检测方法，获得前景运动目标；将隶属于一个连通域的运动目标用矩形框标记出来，提取该矩形框运动目标特征向量获得特征向量集，提取运动目标特征向量，同一个子区域中具有相同人数的目标特征向量组成特征向量集；基于特征向量集建立对应的高斯计数模型，在同一子区域上获得的高斯计数模型组成高斯模型子集，所有子区域上的模型子集组成最终的多高斯计数模型；人流量统计时，设置检测线，对未知人数的视频图像序列进行图像预处理和运动目标检测；对与检测线相交的运动目标区域进行目标跟踪，判断运动目标外接矩形框是否与检测线相交，若不相交，则对下一帧图像进行处理，直到其外接矩形框到达检测线为止，若相交，提取当前运动目标的特征向量；在跟踪过程的每一帧图像中提取当前运动目标的特征向量，根据当前运动目标所处的子区域，采用对应的高斯模型子集分析当前运动目标中的人数，采用快速目标跟踪关联方法获得该目标区域对应的计数队列，并将得到的人数存入该目标区域对应队列；当运动目标离开检测线时，计算队列中人数的平均值，得到人流量统计。In actual detection, it is often only interested in a certain area in the scene, so the region of interest (ROI) is selected first, and all subsequent image processing operations are completed within the region of interest. Divide the region of interest into multiple detection sub-regions of equal size, and use the moving object detection method of making a difference between the foreground image and the background image to obtain the foreground moving object; mark the moving object belonging to a connected domain with a rectangular frame, and extract The moving target eigenvector of the rectangular frame obtains the eigenvector set, extracts the moving target eigenvector, and the target eigenvectors with the same number of people in the same sub-area form the eigenvector set; establishes a corresponding Gaussian counting model based on the eigenvector set, and in the same sub-area The obtained Gaussian counting model constitutes the Gaussian model subset, and the model subsets on all sub-regions constitute the final multi-Gaussian counting model; when counting people, set the detection line, and perform image preprocessing and moving targets on the video image sequence of unknown people Detection: Carry out target tracking on the moving target area intersected with the detection line, judge whether the circumscribed rectangle of the moving target intersects with the detection line, if not, process the next frame of image until the circumscribed rectangle reaches the detection line, If it intersects, extract the feature vector of the current moving target; extract the feature vector of the current moving target in each frame of the tracking process, and use the corresponding Gaussian model subset to analyze the current moving target according to the sub-area where the current moving target is located. number of people, use the fast target tracking association method to obtain the counting queue corresponding to the target area, and store the obtained number of people in the corresponding queue of the target area; when the moving target leaves the detection line, calculate the average number of people in the queue to get the flow of people statistics.

将场景中兴趣区域分割为一系列检测子区域，如可分割为                                                

Figure 190410DEST_PATH_IMAGE001
个面积相同的检测子区域，采用
Figure 175684DEST_PATH_IMAGE002
表示第
Figure 672393DEST_PATH_IMAGE003
行，
Figure 743117DEST_PATH_IMAGE004
列的子区域 (
Figure 734207DEST_PATH_IMAGE005
,
Figure 573987DEST_PATH_IMAGE006
)，其中
Figure 54647DEST_PATH_IMAGE007
,
Figure 47222DEST_PATH_IMAGE008
根据检测视场的大小，以及视场中行人大小，确认其值。一般情况下，检测子区域面积大小可为检测视场中三到四个行人面积大小。对分割后的视频图像进行平滑滤波处理，减少噪声的影响。Divide the region of interest in the scene into a series of detection sub-regions, such as
Figure 190410DEST_PATH_IMAGE001
detection sub-regions with the same area, using
Figure 175684DEST_PATH_IMAGE002
Indicates the first
Figure 672393DEST_PATH_IMAGE003
OK,
Figure 743117DEST_PATH_IMAGE004
subrange of columns (
Figure 734207DEST_PATH_IMAGE005
,
Figure 573987DEST_PATH_IMAGE006
),in
Figure 54647DEST_PATH_IMAGE007
,
Figure 47222DEST_PATH_IMAGE008
Confirm its value according to the size of the detection field of view and the size of pedestrians in the field of view. Generally, the size of the detection sub-region can be the size of three to four pedestrians in the detection field of view. The segmented video image is smoothed and filtered to reduce the influence of noise.

提取背景图像，对检测子区域中当前帧图像和背景图像进行差分处理，得到差分图像。The background image is extracted, and the difference processing is performed on the current frame image and the background image in the detection sub-region to obtain the difference image.

计算差分图像中不同灰度区间像素值的标准方差，并根据这些标准方差中的最大值确定分割阈值，对图像进行分割和二值图像形态学处理，提取前景运动目标。Calculate the standard deviation of pixel values in different gray scale intervals in the difference image, and determine the segmentation threshold according to the maximum value of these standard deviations, perform segmentation and binary image morphology processing on the image, and extract the foreground moving target.

将隶属于一个连通域的运动目标用其外接矩形框标记出来，并提取该运动目标区域的特征向量。Mark the moving target belonging to a connected domain with its circumscribed rectangular frame, and extract the feature vector of the moving target area .

根据矩形框中心点所处的位置，确定运动目标第

Figure 536289DEST_PATH_IMAGE010
次出现所在的检测子区域为，行人个数为
Figure 420118DEST_PATH_IMAGE011
时的特征向量
Figure 815327DEST_PATH_IMAGE012
（本发明实施例中包含运动目标面积为
Figure 301803DEST_PATH_IMAGE013
（像素个数），矩形框长
Figure 389845DEST_PATH_IMAGE014
、宽
Figure 345294DEST_PATH_IMAGE015
三个特征），将具有相同行人个数的特征向量组成特征向量集合
Figure 278615DEST_PATH_IMAGE016
。基于特征向量集建立对应的高斯计数模型
Figure 885176DEST_PATH_IMAGE017
。具体为：According to the position of the center point of the rectangular frame, determine the first moving target
Figure 536289DEST_PATH_IMAGE010
The detection sub-region where the first occurrence is , the number of pedestrians is
Figure 420118DEST_PATH_IMAGE011
eigenvectors when
Figure 815327DEST_PATH_IMAGE012
(Included in the embodiment of the present invention, the area of the moving target is
Figure 301803DEST_PATH_IMAGE013
(number of pixels), the length of the rectangle
Figure 389845DEST_PATH_IMAGE014
,Width
Figure 345294DEST_PATH_IMAGE015
three features), the feature vectors with the same number of pedestrians form a set of feature vectors
Figure 278615DEST_PATH_IMAGE016
. Establish the corresponding Gaussian counting model based on the eigenvector set
Figure 885176DEST_PATH_IMAGE017
. Specifically:

采用公式：

Figure 144119DEST_PATH_IMAGE018
（其中
Figure 632870DEST_PATH_IMAGE008
表示检测子区域为
Figure 822411DEST_PATH_IMAGE002
，检测人数为
Figure 345797DEST_PATH_IMAGE011
时，对应的样本特征向量个数），计算特征向量集
Figure 447745DEST_PATH_IMAGE016
的均值向量，根据均值向量采用公式：
Figure 918489DEST_PATH_IMAGE020
计算其协方差矩阵
Figure 296380DEST_PATH_IMAGE021
，根据公式：
Figure 834809DEST_PATH_IMAGE022
，建立检测子区域为
Figure 298151DEST_PATH_IMAGE002
，检测行人个数为
Figure 829496DEST_PATH_IMAGE011
时，对应的高斯计数模型为
Figure 327473DEST_PATH_IMAGE017
。Using the formula:
Figure 144119DEST_PATH_IMAGE018
(in
Figure 632870DEST_PATH_IMAGE008
Indicates that the detection sub-region is
Figure 822411DEST_PATH_IMAGE002
, the detection number is
Figure 345797DEST_PATH_IMAGE011
, corresponding to the number of sample eigenvectors), calculate the eigenvector set
Figure 447745DEST_PATH_IMAGE016
The mean vector of , according to the mean vector using the formula:
Figure 918489DEST_PATH_IMAGE020
Compute its covariance matrix
Figure 296380DEST_PATH_IMAGE021
, according to the formula:
Figure 834809DEST_PATH_IMAGE022
, establish the detection sub-region as
Figure 298151DEST_PATH_IMAGE002
, the number of detected pedestrians is
Figure 829496DEST_PATH_IMAGE011
When , the corresponding Gaussian counting model is
Figure 327473DEST_PATH_IMAGE017
.

将对应相同检测子区域的高斯计数模型

Figure 771224DEST_PATH_IMAGE017
组成高斯计数模型子集
Figure 987442DEST_PATH_IMAGE023
，将多个高斯计数模型子集
Figure 869947DEST_PATH_IMAGE023
组成多高斯计数模型
Figure 910846DEST_PATH_IMAGE024
。will correspond to a Gaussian counting model of the same detection subregion
Figure 771224DEST_PATH_IMAGE017
Composing a subset of Gaussian count models
Figure 987442DEST_PATH_IMAGE023
, subset multiple Gaussian count models
Figure 869947DEST_PATH_IMAGE023
Composing multi-Gaussian counting models
Figure 910846DEST_PATH_IMAGE024
.

进行人流量统计时，设置检测线，对未知人数的视频图像序列进行图像预处理和运动目标检测，判断运动目标外接矩形框是否与检测线相交，若不相交，则对下一帧图像进行处理，直到其外接矩形框到达检测线为止，若相交，则提取当前运动目标的特征向量，根据其所处的子区域

Figure 587815DEST_PATH_IMAGE002
，调用对应的高斯计数模型子集计算当前运动目标中包含的行人数，并将得到的人数存入该运动目标对应的队列；继续跟踪运动目标，在后续帧中根据其所处的子区域调用对应的高斯计数模型子集，计算该运动目标包含的行人数并存入对应的队列；当运动目标离开检测线，进行人流量统计，即，计算队列中人数的均值，得到当前跟踪目标的人数。When performing people flow statistics, set the detection line, perform image preprocessing and moving target detection on the video image sequence of unknown people, and judge whether the circumscribed rectangular frame of the moving target intersects with the detection line, if not, process the next frame of image , until its circumscribed rectangular frame reaches the detection line, if it intersects, then extract the feature vector of the current moving target, according to the sub-area it is in
Figure 587815DEST_PATH_IMAGE002
, calling the corresponding Gaussian count model subset Calculate the number of pedestrians contained in the current moving target, and store the obtained number in the queue corresponding to the moving target; continue to track the moving target, and call the corresponding Gaussian counting model subset in subsequent frames according to the sub-region where it is located, and calculate The number of pedestrians included in the moving target is stored in the corresponding queue; when the moving target leaves the detection line, the flow of people is counted, that is, the average value of the number of people in the queue is calculated to obtain the number of people currently tracking the target.

本发明通过对兴趣区域中的运动目标检测，建立高斯计数模型进行运动目标人数计算，能够在大密度的人流量时进行较高精度的人流量统计，并降低了计算复杂度，达到实时性要求。The invention detects the moving target in the interest area, establishes a Gaussian counting model to calculate the number of moving targets, can perform high-precision counting of the flow of people in the case of a large-density flow of people, and reduces the computational complexity to meet real-time requirements .

附图说明Description of drawings

图1本发明实施例中基于多高斯计数模型的人流量统计方法的流程图；Fig. 1 is the flow chart of the people flow counting method based on multi-Gaussian counting model in the embodiment of the present invention;

图2本发明实施例中多高斯计数模型建立阶段流程图；Figure 2 is a flow chart of the establishment stage of the multi-Gaussian counting model in the embodiment of the present invention;

图3本发明实施例中前景运动目标检测流程图；Figure 3 is a flow chart of foreground moving target detection in an embodiment of the present invention;

图4本发明实施例中目标跟踪示意图。Fig. 4 is a schematic diagram of target tracking in an embodiment of the present invention.

具体实施方式Detailed ways

下面结合附图和具体实施例对本发明作详细描述。The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

本发明提出的一种基于多高斯计数模型的人流量统计用于实时监控系统。通过运动检测获得前景运动目标，根据多高斯计数模型分析当前图像中的人数，实现人流量统计。A kind of people flow statistics based on multi-Gauss counting model proposed by the present invention is used in real-time monitoring system. The foreground moving target is obtained through motion detection, and the number of people in the current image is analyzed according to the multi-Gaussian counting model to realize people flow statistics.

图1为本发明实施例中基于多高斯计数模型的人流量统计流程图。如图1所示，本发明实施例中在进行人流量统计前需要建立多高斯计数模型。FIG. 1 is a flow chart of people flow statistics based on a multi-Gaussian counting model in an embodiment of the present invention. As shown in FIG. 1 , in the embodiment of the present invention, a multi-Gaussian counting model needs to be established before performing people flow statistics.

将场景中兴趣区域分割为一系列检测子区域。对分割后的视频图像进行平滑滤波处理。Segment the region of interest in the scene into a series of detection sub-regions. Smoothing and filtering are performed on the segmented video images.

提取背景图像，对检测子区域中当前帧图像和背景图像进行差分处理，得到差分图像。The background image is extracted, and the difference processing is performed on the current frame image and the background image in the detection sub-region to obtain the difference image.

计算差分图像中不同灰度区间像素值的标准方差，并根据这些标准方差中的最大值确定分割阈值，对图像进行分割和二值图像形态学处理，提取前景运动目标。Calculate the standard deviation of pixel values in different gray scale intervals in the difference image, and determine the segmentation threshold according to the maximum value of these standard deviations, perform segmentation and binary image morphology processing on the image, and extract the foreground moving target.

将隶属于一个连通域的运动目标用其外接矩形框标记出来，并提取该运动目标区域的特征向量

Figure 649629DEST_PATH_IMAGE009
。Mark the moving target belonging to a connected domain with its circumscribed rectangular frame, and extract the feature vector of the moving target area
Figure 649629DEST_PATH_IMAGE009
.

根据矩形框中心点所处的位置，确定运动目标第

Figure 43570DEST_PATH_IMAGE010
次出现所在的检测子区域为
Figure 157020DEST_PATH_IMAGE002
，行人个数为
Figure 19934DEST_PATH_IMAGE011
时的特征向量
Figure 244242DEST_PATH_IMAGE012
（本发明实施例中特征向量包含运动目标面积为
Figure 259733DEST_PATH_IMAGE013
（像素个数），矩形框长
Figure 544084DEST_PATH_IMAGE014
、宽
Figure 628715DEST_PATH_IMAGE015
三个特征），将具有相同行人个数的特征向量组成特征向量集合
Figure 656714DEST_PATH_IMAGE016
，基于特征向量集建立对应的高斯计数模型
Figure 25247DEST_PATH_IMAGE017
。具体为：According to the position of the center point of the rectangular frame, determine the first moving target
Figure 43570DEST_PATH_IMAGE010
The detection sub-region where the first occurrence is
Figure 157020DEST_PATH_IMAGE002
, the number of pedestrians is
Figure 19934DEST_PATH_IMAGE011
eigenvectors when
Figure 244242DEST_PATH_IMAGE012
(In the embodiment of the present invention, the feature vector includes the area of the moving target as
Figure 259733DEST_PATH_IMAGE013
(number of pixels), the length of the rectangle
Figure 544084DEST_PATH_IMAGE014
,Width
Figure 628715DEST_PATH_IMAGE015
three features), the feature vectors with the same number of pedestrians form a set of feature vectors
Figure 656714DEST_PATH_IMAGE016
, based on the eigenvector set to establish the corresponding Gaussian counting model
Figure 25247DEST_PATH_IMAGE017
. Specifically:

采用公式：（其中

Figure 318005DEST_PATH_IMAGE008
表示检测子区域为，检测人数为
Figure 608620DEST_PATH_IMAGE011
时，对应的样本特征向量个数），计算特征向量集
Figure 172457DEST_PATH_IMAGE016
的均值向量
Figure 293997DEST_PATH_IMAGE019
，根据均值向量采用公式：
Figure 850749DEST_PATH_IMAGE020
计算其协方差矩阵
Figure 6924DEST_PATH_IMAGE021
，根据公式：
Figure 741661DEST_PATH_IMAGE022
，建立检测子区域为
Figure 350497DEST_PATH_IMAGE002
，检测行人个数为
Figure 946826DEST_PATH_IMAGE011
时，对应的高斯计数模型为
Figure 895190DEST_PATH_IMAGE017
。Using the formula: (in
Figure 318005DEST_PATH_IMAGE008
Indicates that the detection sub-region is , the detection number is
Figure 608620DEST_PATH_IMAGE011
, corresponding to the number of sample eigenvectors), calculate the eigenvector set
Figure 172457DEST_PATH_IMAGE016
The mean vector of
Figure 293997DEST_PATH_IMAGE019
, according to the mean vector using the formula:
Figure 850749DEST_PATH_IMAGE020
Compute its covariance matrix
Figure 6924DEST_PATH_IMAGE021
, according to the formula:
Figure 741661DEST_PATH_IMAGE022
, establish the detection sub-region as
Figure 350497DEST_PATH_IMAGE002
, the number of detected pedestrians is
Figure 946826DEST_PATH_IMAGE011
When , the corresponding Gaussian counting model is
Figure 895190DEST_PATH_IMAGE017
.

将对应相同检测子区域的高斯计数模型组成高斯计数模型子集

Figure 146229DEST_PATH_IMAGE023
，将多个高斯计数模型子集
Figure 795516DEST_PATH_IMAGE023
组成多高斯计数模型
Figure 926283DEST_PATH_IMAGE024
。will correspond to a Gaussian counting model of the same detection subregion Composing a subset of Gaussian count models
Figure 146229DEST_PATH_IMAGE023
, subset multiple Gaussian count models
Figure 795516DEST_PATH_IMAGE023
Composing multi-Gaussian counting models
Figure 926283DEST_PATH_IMAGE024
.

进行人流量统计时，设置检测线，对未知人数的视频图像序列进行图像预处理和运动目标检测，判断运动目标外接矩形框是否与检测线相交，若不相交，则对下一帧图像进行处理，直到其外接矩形框到达检测线为止，若相交，则提取当前运动目标的特征向量，根据其所处的子区域

Figure 65141DEST_PATH_IMAGE002
，调用对应的高斯计数模型子集
Figure 71405DEST_PATH_IMAGE023
计算当前运动目标中包含的行人数，并将得到的人数存入该运动目标对应的队列；继续跟踪运动目标，在后续帧中根据其所处的子区域调用对应的高斯计数模型子集，计算该运动目标包含的行人数并存入对应的队列；当运动目标离开检测线，进行人流量统计，即，计算队列中人数的均值，得到当前跟踪目标的人数。When performing people flow statistics, set the detection line, perform image preprocessing and moving target detection on the video image sequence of unknown people, and judge whether the circumscribed rectangular frame of the moving target intersects with the detection line, if not, process the next frame of image , until its circumscribed rectangular frame reaches the detection line, if it intersects, then extract the feature vector of the current moving target, according to the sub-area it is in
Figure 65141DEST_PATH_IMAGE002
, calling the corresponding Gaussian count model subset
Figure 71405DEST_PATH_IMAGE023
Calculate the number of pedestrians contained in the current moving target, and store the obtained number in the queue corresponding to the moving target; continue to track the moving target, and call the corresponding Gaussian counting model subset in subsequent frames according to the sub-region where it is located, and calculate The number of pedestrians included in the moving target is stored in the corresponding queue; when the moving target leaves the detection line, the flow of people is counted, that is, the average value of the number of people in the queue is calculated to obtain the number of people currently tracking the target.

图2为本发明实施例中多高斯计数模型建立流程图，具体包括如下几个步骤：Fig. 2 is a flow chart for establishing a multi-Gaussian counting model in an embodiment of the present invention, which specifically includes the following steps:

通过现场拍摄获得带人数标记的视频图像序列。A sequence of video images marked with people is obtained through on-site shooting.

对检测区域进行划分。选择输入的视频图像的兴趣区域，然后将兴趣区域划分为一系列检测子区域，将场景中兴趣区域分割为一系列检测子区域，如可分割为

Figure 586700DEST_PATH_IMAGE001
个面积相同的检测子区域，采用表示第
Figure 819415DEST_PATH_IMAGE025
行，
Figure 811511DEST_PATH_IMAGE004
列的子区域 (,)，其中
Figure 123041DEST_PATH_IMAGE007
,
Figure 681061DEST_PATH_IMAGE008
根据检测视场的大小，以及视场中行人大小，确认其值。一般情况下，检测子区域面积大小可为检测视场中三到四个行人面积大小。对分割后的视频图像进行平滑滤波处理，减少噪声的影响。Divide the detection area. Select the region of interest of the input video image, and then divide the region of interest into a series of detection sub-regions, and divide the region of interest in the scene into a series of detection sub-regions, such as
Figure 586700DEST_PATH_IMAGE001
detection sub-regions with the same area, using Indicates the first
Figure 819415DEST_PATH_IMAGE025
OK,
Figure 811511DEST_PATH_IMAGE004
subrange of columns ( , ),in
Figure 123041DEST_PATH_IMAGE007
,
Figure 681061DEST_PATH_IMAGE008
Confirm its value according to the size of the detection field of view and the size of pedestrians in the field of view. Generally, the size of the detection sub-region can be the size of three to four pedestrians in the detection field of view. The segmented video image is smoothed and filtered to reduce the influence of noise.

图像预处理。Image preprocessing.

将兴趣区域的输入视频序列图像转化为灰度图像序列，预处理模块对灰度图像序列进行滤波，消除图像中的噪声。本发明实施例采用高斯平滑对图像进行滤波。The input video sequence image of the region of interest is converted into a grayscale image sequence, and the preprocessing module filters the grayscale image sequence to eliminate noise in the image. In the embodiment of the present invention, Gaussian smoothing is used to filter the image.

运动目标检测。提取背景图像：根据监控场景的特点，选择合适的背景图像提取。将当前帧图像与背景图像进行差分运算获得差分图像，进行图像阈值分割和二值图像形态学处理，获得前景运动目标。Moving object detection. Extract background image: According to the characteristics of the monitoring scene, select the appropriate background image to extract. The differential operation is performed on the current frame image and the background image to obtain the difference image, and image threshold segmentation and binary image morphology processing are performed to obtain the foreground moving target.

特征向量提取。Feature vector extraction.

获取矩形框中运动目标面积，以及矩形框的长度和宽度。以运动目标面积为(

Figure 949276DEST_PATH_IMAGE013
)，矩形框长(
Figure 909142DEST_PATH_IMAGE014
)、宽(
Figure 498386DEST_PATH_IMAGE015
)三个特征构造特征向量，即
Figure 278123DEST_PATH_IMAGE026
。表示第
Figure 674655DEST_PATH_IMAGE010
次出现的运动目标所在的子区域为
Figure 497118DEST_PATH_IMAGE028
，行人个数为时的特征向量；Obtain the area of the moving target in the rectangular frame, as well as the length and width of the rectangular frame. Taking the target area of the movement as (
Figure 949276DEST_PATH_IMAGE013
), the length of the rectangle (
Figure 909142DEST_PATH_IMAGE014
),Width(
Figure 498386DEST_PATH_IMAGE015
) three features to construct the feature vector, namely
Figure 278123DEST_PATH_IMAGE026
. Indicates the first
Figure 674655DEST_PATH_IMAGE010
The sub-area where the moving target appears for the first time is
Figure 497118DEST_PATH_IMAGE028
, the number of pedestrians is eigenvector when

将同一个子区域内人数相同的目标连通域（即具有相同

Figure 900734DEST_PATH_IMAGE029
）的特征向量
Figure 258029DEST_PATH_IMAGE030
组成特征向量集合
Figure 251392DEST_PATH_IMAGE016
。The target connected domain with the same number of people in the same sub-area (that is, with the same
Figure 900734DEST_PATH_IMAGE029
) eigenvector
Figure 258029DEST_PATH_IMAGE030
Make up the set of eigenvectors
Figure 251392DEST_PATH_IMAGE016
.

分析各个特征向量集中向量个数，如向量个数小于

Figure 943405DEST_PATH_IMAGE008
，则继续提取特征向量，其中
Figure 945996DEST_PATH_IMAGE008
的取值越大越好，根据经验一般取
Figure 469381DEST_PATH_IMAGE031
。当特征向量集中向量个数大于等于
Figure 820597DEST_PATH_IMAGE008
，根据特征向量集，建立单高斯计数模型
Figure 62223DEST_PATH_IMAGE017
。单高斯计数模型
Figure 540608DEST_PATH_IMAGE017
的建立过程如下：Analyze the number of vectors in each feature vector set, if the number of vectors is less than
Figure 943405DEST_PATH_IMAGE008
, then continue to extract the feature vector, where
Figure 945996DEST_PATH_IMAGE008
The larger the value of the better, according to experience generally take
Figure 469381DEST_PATH_IMAGE031
. When the number of vectors in the eigenvector set is greater than or equal to
Figure 820597DEST_PATH_IMAGE008
, according to the eigenvector set, a single Gaussian count model is established
Figure 62223DEST_PATH_IMAGE017
. Single Gaussian Count Model
Figure 540608DEST_PATH_IMAGE017
The establishment process is as follows:

调用公式：

Figure 918500DEST_PATH_IMAGE018
计算特征向量集
Figure 207661DEST_PATH_IMAGE016
的均值向量
Figure 671004DEST_PATH_IMAGE019
，根据均值向量利用公式：
Figure 953080DEST_PATH_IMAGE020
计算其协方差矩阵，根据协方差矩阵建立检测区域为
Figure 144076DEST_PATH_IMAGE002
，检测行人个数为
Figure 94715DEST_PATH_IMAGE011
时，对应的高斯计数模型:
Figure 267387DEST_PATH_IMAGE022
,式中，为随机向量，
Figure 336285DEST_PATH_IMAGE033
表示转置。Call formula:
Figure 918500DEST_PATH_IMAGE018
Calculate the set of eigenvectors
Figure 207661DEST_PATH_IMAGE016
The mean vector of
Figure 671004DEST_PATH_IMAGE019
, according to the mean vector using the formula:
Figure 953080DEST_PATH_IMAGE020
Compute its covariance matrix , according to the covariance matrix to establish the detection area as
Figure 144076DEST_PATH_IMAGE002
, the number of detected pedestrians is
Figure 94715DEST_PATH_IMAGE011
, the corresponding Gaussian counting model :
Figure 267387DEST_PATH_IMAGE022
, where, is a random vector,
Figure 336285DEST_PATH_IMAGE033
Indicates a transpose.

当对所有的向量集建立对应的高斯计数模型后，处于同一个子区域（即具有相同

Figure 167155DEST_PATH_IMAGE034
）的高斯计数模型
Figure 467555DEST_PATH_IMAGE017
组成高斯计数模型子集，所有的高斯计数模型子集组成多高斯计数模型
Figure 881853DEST_PATH_IMAGE024
，即
Figure 900625DEST_PATH_IMAGE035
。When for all vector sets After establishing the corresponding Gaussian counting model, they are in the same sub-region (that is, have the same
Figure 167155DEST_PATH_IMAGE034
) Gaussian count model
Figure 467555DEST_PATH_IMAGE017
Composing a subset of Gaussian count models , a subset of all Gaussian count models Composing multi-Gaussian counting models
Figure 881853DEST_PATH_IMAGE024
,Right now
Figure 900625DEST_PATH_IMAGE035
.

完成多高斯计数模型建立后，可以根据该计数模型进行人流量统计。如图1所示，具体包括如下步骤：After the multi-Gaussian counting model is established, people flow statistics can be performed according to the counting model. As shown in Figure 1, it specifically includes the following steps:

通过单个CCD成像传感器通过垂直拍摄获得视频序列图像。对检测区域进行区域划分。Video sequence images are obtained by vertical shooting through a single CCD imaging sensor. Divide the detection area.

设置检测线，最好选择兴趣区域的中线，如图4的线段L1。对图像进行预处理，运动目标检测。To set the detection line, it is best to choose the midline of the region of interest, such as line segment L1 in Figure 4. Image preprocessing, moving target detection.

若未检测到运动目标，则转至下一帧继续处理；若检测到运动目标，则判断运动目标外接矩形框是否与检测线相交，若不相交，则转至下一帧继续处理，否则提取当前运动目标的特征向量

Figure 1567DEST_PATH_IMAGE036
（与多高斯计数模型建立阶段的特征向量一样）。估计当前运动目标区域包含的人数，具体包括如下步骤：If no moving object is detected, go to the next frame to continue processing; if a moving object is detected, then judge whether the circumscribed rectangle of the moving object intersects with the detection line, if not, go to the next frame to continue processing, otherwise extract The eigenvector of the current moving target
Figure 1567DEST_PATH_IMAGE036
(Same as the eigenvectors of the multi-Gaussian count model building phase). Estimate the number of people contained in the current motion target area, specifically including the following steps:

根据运动目标外接矩形框中心点坐标确定其所处的子区域

Figure 29566DEST_PATH_IMAGE002
，在多高斯计数模型中获得子区域
Figure 148832DEST_PATH_IMAGE002
对应的高斯计数模型子集
Figure 604084DEST_PATH_IMAGE023
，即
Figure 425278DEST_PATH_IMAGE023
为
Figure 991389DEST_PATH_IMAGE037
，
Figure 230740DEST_PATH_IMAGE038
，
Figure 856894DEST_PATH_IMAGE039
，
Figure 978433DEST_PATH_IMAGE040
组成的集合。把当前特征向量
Figure 36650DEST_PATH_IMAGE036
分别带入模型(高斯函数)
Figure 192825DEST_PATH_IMAGE037
，，
Figure 536399DEST_PATH_IMAGE039
，
Figure 631263DEST_PATH_IMAGE040
 中计算，则计算结果为最大值的模型对应的人数为当前运动目标区域包含的人数，其值等于
Figure 609900DEST_PATH_IMAGE011
。即：According to the coordinates of the center point of the circumscribed rectangular frame of the moving target, determine its sub-area
Figure 29566DEST_PATH_IMAGE002
, to obtain subregions in a multi-Gaussian counting model
Figure 148832DEST_PATH_IMAGE002
The corresponding subset of Gaussian count models
Figure 604084DEST_PATH_IMAGE023
,Right now
Figure 425278DEST_PATH_IMAGE023
for
Figure 991389DEST_PATH_IMAGE037
,
Figure 230740DEST_PATH_IMAGE038
,
Figure 856894DEST_PATH_IMAGE039
,
Figure 978433DEST_PATH_IMAGE040
composed of collections. put the current eigenvector
Figure 36650DEST_PATH_IMAGE036
Bring into the model (Gaussian function) respectively
Figure 192825DEST_PATH_IMAGE037
, ,
Figure 536399DEST_PATH_IMAGE039
,
Figure 631263DEST_PATH_IMAGE040
Calculated in, the calculation result is the model of the maximum value The corresponding number of people is the number of people contained in the current exercise target area, and its value is equal to
Figure 609900DEST_PATH_IMAGE011
. Right now:

Figure 643715DEST_PATH_IMAGE041
   
Figure 355319DEST_PATH_IMAGE042
Figure 643715DEST_PATH_IMAGE041
   
Figure 355319DEST_PATH_IMAGE042

计算当前运动目标中包含的人数

Figure 908923DEST_PATH_IMAGE043
。其中，
Figure 47780DEST_PATH_IMAGE044
表示模型子集中的最大值。Calculate the number of people included in the current exercise goal
Figure 908923DEST_PATH_IMAGE043
. in,
Figure 47780DEST_PATH_IMAGE044
Represents a subset of models the maximum value in .

获得运动目标区域包含的人数后，可通过基于检测线的快速跟踪方法，获得该目标区域对应的计数队列，并把当前获得人数估计存入该计数队列中。将统计人数记录到该检测目标对应的队列中。After obtaining the number of people contained in the moving target area, the counting queue corresponding to the target area can be obtained through the fast tracking method based on the detection line, and the currently obtained number of people is estimated to be stored in the counting queue. Record the number of statistics to the queue corresponding to the detection target.

如图3所示为前景运动目标检测流程图，具体包括如下步骤：As shown in Figure 3, it is a flow chart of foreground moving target detection, which specifically includes the following steps:

提取背景图像：根据监控场景的特点，选择合适的背景图像提取方法。如：对于一些背景不变化或者变化很小的特殊环境（如部分室内环境），直接拍摄背景图像，而后保持背景图像不变。对于背景变化明显的场景（如自然场景），使用基于直方图的背景建模法得到背景图像，具体为：统计序列图像在同一像素点位置上像素值的灰度直方图，出现次数最多的灰度值作为该点的背景像素值。Extract the background image: According to the characteristics of the monitoring scene, select the appropriate background image extraction method. For example, for some special environments (such as some indoor environments) where the background does not change or changes little, directly shoot the background image, and then keep the background image unchanged. For scenes with obvious background changes (such as natural scenes), use the histogram-based background modeling method to obtain the background image, specifically: the gray histogram of the pixel values at the same pixel position in the statistical sequence image, the gray value with the most occurrences degree value as the background pixel value of the point.

获得差分图像：运动目标检测处理模块中，把从视频图像序列中获取的当前图像帧与背景图像进行差分运算，获得差分图像。Obtaining the difference image: In the moving object detection processing module, the current image frame obtained from the video image sequence and the background image are subjected to a difference operation to obtain a difference image.

差分图像阈值分割：获取的差分图像为标准的8位灰度图像，因此其像素值范围为[0,255]，依次求取差分图像中灰度范围分别为[0,255]、[1,255]、……、[254,255]，总共254个灰度区间内像素值的标准方差，然后计算这个254个标准方差的最大值

Figure 84186DEST_PATH_IMAGE046
，并设置两个阈值
Figure 803880DEST_PATH_IMAGE047
，
Figure 300590DEST_PATH_IMAGE048
(
Figure 371314DEST_PATH_IMAGE049
)，（
Figure 362404DEST_PATH_IMAGE047
、
Figure 202184DEST_PATH_IMAGE048
可根据输入视频图像质量选取，一般情况下，
Figure 371259DEST_PATH_IMAGE050
，
Figure 929279DEST_PATH_IMAGE051
），根据
Figure 520798DEST_PATH_IMAGE047
、
Figure 152767DEST_PATH_IMAGE048
确定中间阈值T，通过如下公式求取最终的分割阈值
Figure 69908DEST_PATH_IMAGE052
；Differential image threshold segmentation: The obtained differential image is a standard 8-bit grayscale image, so its pixel value range is [0,255], and the grayscale ranges in the differential image are respectively [0,255], [1,255],..., [254,255], the standard deviation of pixel values in a total of 254 grayscale intervals, and then calculate the maximum value of the 254 standard deviations
Figure 84186DEST_PATH_IMAGE046
, and set two thresholds
Figure 803880DEST_PATH_IMAGE047
,
Figure 300590DEST_PATH_IMAGE048
(
Figure 371314DEST_PATH_IMAGE049
), (
Figure 362404DEST_PATH_IMAGE047
,
Figure 202184DEST_PATH_IMAGE048
It can be selected according to the quality of the input video image. In general,
Figure 371259DEST_PATH_IMAGE050
,
Figure 929279DEST_PATH_IMAGE051
),according to
Figure 520798DEST_PATH_IMAGE047
,
Figure 152767DEST_PATH_IMAGE048
Determine the intermediate threshold T, and obtain the final segmentation threshold by the following formula
Figure 69908DEST_PATH_IMAGE052
;

Figure 36596DEST_PATH_IMAGE053
，
Figure 36596DEST_PATH_IMAGE053
,

， ,

其中

Figure 183860DEST_PATH_IMAGE044
、分别表示求取两个数值的最大值和最小值。以
Figure 273356DEST_PATH_IMAGE052
作为阈值，对图像进行分割，得到二值图像。in
Figure 183860DEST_PATH_IMAGE044
, Respectively means to find the maximum value and minimum value of two values. by
Figure 273356DEST_PATH_IMAGE052
As a threshold, the image is segmented to obtain a binary image.

二值图像形态学处理：在二值图像中进行形态学滤波，即通过形态学腐蚀操作删除一些面积较小的虚假目标区域，并通过膨胀操作把一些断裂的目标区域进行区域合并。然后，采用8近邻连通域搜索算法搜索各个连通区域，用连通域的外接矩形进行标记，由此获得前景运动目标。Binary image morphology processing: Morphological filtering is performed in binary images, that is, some small false target areas are deleted through morphological erosion operations, and some broken target areas are merged through expansion operations. Then, the 8-nearest neighbor connected domain search algorithm is used to search each connected area, and the circumscribed rectangle of the connected area is used to mark the foreground moving target.

目标快速跟踪方法如图4所示，具体包括如下步骤：本实施例中目标跟踪的目的就是实现连续两帧图像中运动目标区域的关联，即判断前后两帧图像中与检测线相交的运动目标是否为同一个目标。The fast target tracking method is shown in Figure 4, and specifically includes the following steps: the purpose of target tracking in this embodiment is to realize the association of moving target areas in two consecutive frames of images, that is, to determine the moving targets intersecting the detection line in the two frames of images before and after whether it is the same target.

图4中展示了单个运动目标和前后距离较近的两个目标情况下，前后两帧图像中的状态。图中，区域R1为视频输入图像；区域R2为兴趣区域（可以根据实际需要选择适当的形状以及大小的兴趣区域）；区域R3为运动目标的外接矩形框；线段L1为检测线，一般选择兴趣区域的中线；线段L2为运动目标与检测线的交线段；P1，P2为前后两帧（第帧和第

Figure 172390DEST_PATH_IMAGE056
帧）图像中外接矩形框的右上角的顶点。Figure 4 shows the status of the two frames of images before and after a single moving target and two targets with a short distance between the front and back. In the figure, the region R1 is the video input image; the region R2 is the region of interest (you can choose the appropriate shape and size of the region of interest according to actual needs); the region R3 is the circumscribed rectangular frame of the moving target; the line segment L1 is the detection line, generally select the interest region The center line of the area; the line segment L2 is the intersection segment between the moving target and the detection line; P1 and P2 are the two frames before and after (frame and frame
Figure 172390DEST_PATH_IMAGE056
Frame) is the top-right vertex of the bounding rectangle in the image.

假设P1和P2的坐标分别为

Figure 841269DEST_PATH_IMAGE057
和
Figure 772316DEST_PATH_IMAGE058
（图像坐标系，即图像左上角为原点，X轴正方向为水平向右，Y轴正方向为垂直向下），通过两个约束条件来判别前后两帧图像中的运动目标是否为同一个目标，即：（1）前后两帧图像中的运动目标区域与检测线的交线段重叠或者部分重叠；（2）坐标
Figure 526645DEST_PATH_IMAGE059
。若前后两帧图像中的运动目标同时满足（1）和（2）两个约束条件，则判断为同一个目标，否者判断为不同的目标。对于只包含当个运动目标的情况，用约束（1）即可实现正确判断，如图4（a）所示。但对图4（b）的前后距离较近的两个运动目标的情况，尽管前后两帧的交线段重叠，但是必须通过约束（2）才能实现正确判断。由于处理帧频较高，（1）和（2）两个约束判断条件可以实现绝大数情况的正确判断，该算法简单、高效。Suppose the coordinates of P1 and P2 are respectively
Figure 841269DEST_PATH_IMAGE057
and
Figure 772316DEST_PATH_IMAGE058
(The image coordinate system, that is, the upper left corner of the image is the origin, the positive direction of the X-axis is horizontal to the right, and the positive direction of the Y-axis is vertical downward). Two constraints are used to determine whether the moving targets in the two frames of images before and after are the same The target, that is: (1) the moving target area in the two frames of images before and after overlaps or partially overlaps with the intersection of the detection line; (2) the coordinates
Figure 526645DEST_PATH_IMAGE059
. If the moving targets in the two frames before and after satisfying the constraints (1) and (2) at the same time, they are judged as the same target, otherwise they are judged as different targets. For the situation that only includes one moving target, the correct judgment can be realized by using constraint (1), as shown in Figure 4(a). But for the case of two moving targets with close distances in Figure 4(b), although the intersection line segments of the two frames overlap, the correct judgment must be achieved through constraint (2). Due to the high processing frame rate, the two constraint judgment conditions (1) and (2) can realize the correct judgment in most cases, and the algorithm is simple and efficient.

若运动目标离开检测线，即运动目标与检测线连续三帧不相交时，满足计数条件，计算当前运动目标对应的计数队列中行人人数的均值

Figure 450608DEST_PATH_IMAGE060
，得到当前运动目标所包含的人数。然后累加各个队列行人人数的均值，得到总的人流量
Figure 973993DEST_PATH_IMAGE061
，从而实现人流量统计。If the moving object leaves the detection line, that is, when the moving object does not intersect with the detection line for three consecutive frames, the counting condition is met, and the average number of pedestrians in the counting queue corresponding to the current moving object is calculated
Figure 450608DEST_PATH_IMAGE060
, to get the number of people included in the current exercise target. Then add up the mean value of the number of pedestrians in each queue to get the total flow of people
Figure 973993DEST_PATH_IMAGE061
, so as to realize people flow statistics.

以上所述仅是本发明的优选实施方式，应该指出，对于本技术领域的普通技术人员来说，在不脱离本发明原理的前提下，还可以做出若干改进和润饰，这些改进和润饰也应视为本发明的保护范围。The above are only preferred embodiments of the present invention, and it should be pointed out that for those of ordinary skill in the art, some improvements and modifications can also be made without departing from the principles of the present invention. It should be regarded as the protection scope of the present invention.

Claims (7)

Translated fromChinese
1.一种基于多高斯计数模型的人流量统计方法，其特征在于，包括如下步骤，将视频图像分割为检测子区域，将当前帧图像和背景图像做差分，得到差分图像，获得前景运动目标；将隶属于一个连通域的运动目标用矩形框标记出来，提取该矩形框运动目标特征向量获得特征向量集；基于特征向量集                                                
Figure 179095DEST_PATH_IMAGE001
建立对应的高斯计数模型，属于相同子区域的高斯计数模型组成高斯计数模型子集，所有的高斯计数模型子集构成多高斯计数模型；判断运动目标外接矩形框是否与检测线相交，若不相交，则对下一帧图像进行处理，直到其外接矩形框到达检测线为止；若相交，提取当前运动目标的特征向量，根据当前运动目标所处的子区域，用对应的高斯计数模型子集分析当前运动目标人数，采用快速目标跟踪关联方法获得该目标区域对应的计数队列，并将当前运动目标人数存入该队列，当目标离开检测线时，计算队列的平均值得到该目标区域包含的人数。1. A kind of people counting method based on many Gaussian counting models, it is characterized in that, comprise the steps, video image is divided into detection sub-region, current frame image and background image are done difference, obtain difference image, obtain foreground moving target ;Mark the moving target belonging to a connected domain with a rectangular frame, extract the feature vector of the moving target in the rectangular frame to obtain the feature vector set; based on the feature vector set
Figure 179095DEST_PATH_IMAGE001
Establish the corresponding Gaussian counting model, the Gaussian counting models belonging to the same sub-region form the Gaussian counting model subset, and all the Gaussian counting model subsets constitute the multi-Gaussian counting model; determine whether the circumscribed rectangle of the moving target intersects with the detection line, if not , then process the next frame of image until its circumscribed rectangular frame reaches the detection line; if it intersects, extract the feature vector of the current moving target, and analyze it with the corresponding Gaussian counting model subset according to the sub-region where the current moving target is located For the current number of moving targets, the fast target tracking association method is used to obtain the counting queue corresponding to the target area, and the current number of moving targets is stored in the queue. When the target leaves the detection line, the average value of the queue is calculated to obtain the number of people contained in the target area. .2.根据权利要求1所述的人流量统计方法，其特征在于，所述获得的前景运动目标具体为：提取背景图像，计算检测子区域中当前图像与背景图像的差分图像中各灰度区间像素值的标准方差的最大值，并以此确定分割阈值，根据分割阈值对差分图像进行分割，及二值图像形态学处理获得前景运动目标。2. The method for counting people flow according to claim 1, wherein the foreground moving target obtained is specifically: extracting the background image, calculating and detecting each gray-scale interval in the difference image between the current image and the background image in the detection sub-region The maximum value of the standard deviation of the pixel value is used to determine the segmentation threshold, and the difference image is segmented according to the segmentation threshold, and the binary image morphology is processed to obtain the foreground moving target.3.根据权利要求1所述的人流量统计方法，其特征在于，根据矩形框面积、长、宽确定运动目标第
Figure 92824DEST_PATH_IMAGE002
次出现在检测子区域为，行人个数为
Figure 463948DEST_PATH_IMAGE004
时的特征向量
Figure 799115DEST_PATH_IMAGE005
 ，将具有相同行人个数的特征向量组成特征向量集合，基于特征向量集建立对应的高斯计数模型
Figure 544534DEST_PATH_IMAGE006
，并把处于同一个子区域的
Figure 42511DEST_PATH_IMAGE006
归为一个高斯计数模型子集
Figure 236995DEST_PATH_IMAGE007
 。3. The method for counting the flow of people according to claim 1 is characterized in that, according to the rectangular frame area, length and width, it is determined that the moving target is the first
Figure 92824DEST_PATH_IMAGE002
Occurs in the detection sub-region as , the number of pedestrians is
Figure 463948DEST_PATH_IMAGE004
eigenvectors when
Figure 799115DEST_PATH_IMAGE005
, the feature vectors with the same number of pedestrians form a set of feature vectors , based on the eigenvector set to establish the corresponding Gaussian counting model
Figure 544534DEST_PATH_IMAGE006
, and put in the same sub-region
Figure 42511DEST_PATH_IMAGE006
into a subset of Gaussian count models
Figure 236995DEST_PATH_IMAGE007
.4.根据权利要求1所述的人流量统计方法，其特征在于，差分图像阈值分割具体为：获取差分图像的灰度图像，依次求取差分图像中灰度区间内像素值的标准方差，总共标准方差的最大值
Figure 453212DEST_PATH_IMAGE008
，根据公式：
Figure 273401DEST_PATH_IMAGE009
，
Figure 625885DEST_PATH_IMAGE010
求取分割阈值
Figure 755384DEST_PATH_IMAGE011
，以
Figure 193318DEST_PATH_IMAGE011
作为阈值，对图像进行分割，得到二值图像，其中，
Figure 613935DEST_PATH_IMAGE012
，
Figure 24188DEST_PATH_IMAGE013
为 设置的阈值(
Figure 872058DEST_PATH_IMAGE014
)。4. The method for counting people flow according to claim 1, wherein the differential image threshold segmentation is specifically as follows: obtaining the grayscale image of the differential image, and sequentially obtaining the standard deviation of the pixel values in the grayscale interval in the differential image, a total of the maximum value of the standard deviation
Figure 453212DEST_PATH_IMAGE008
, according to the formula:
Figure 273401DEST_PATH_IMAGE009
,
Figure 625885DEST_PATH_IMAGE010
Find the segmentation threshold
Figure 755384DEST_PATH_IMAGE011
,by
Figure 193318DEST_PATH_IMAGE011
As a threshold, the image is segmented to obtain a binary image, where,
Figure 613935DEST_PATH_IMAGE012
,
Figure 24188DEST_PATH_IMAGE013
The threshold set for (
Figure 872058DEST_PATH_IMAGE014
).5.根据权利要求1或3所述的流量统计方法，其特征在于，基于特征向量集
Figure 485704DEST_PATH_IMAGE001
建立对应的高斯计数模型
Figure 710012DEST_PATH_IMAGE006
，具体为：根据特征向量调用公式：
Figure 974772DEST_PATH_IMAGE015
计算特征向量集的均值向量，根据均值向量利用公式：计算协方差矩阵
Figure 802602DEST_PATH_IMAGE018
，根据公式：
Figure 195537DEST_PATH_IMAGE019
建立检测区域为
Figure 95360DEST_PATH_IMAGE020
，检测行人个数为
Figure 349886DEST_PATH_IMAGE004
时，对应的高斯计数模型为
Figure 385976DEST_PATH_IMAGE006
。5. The traffic statistics method according to claim 1 or 3, characterized in that, based on the feature vector set
Figure 485704DEST_PATH_IMAGE001
Build the corresponding Gaussian counting model
Figure 710012DEST_PATH_IMAGE006
, specifically: call the formula according to the eigenvector:
Figure 974772DEST_PATH_IMAGE015
Calculate the set of eigenvectors The mean vector of , according to the mean vector using the formula: Compute the covariance matrix
Figure 802602DEST_PATH_IMAGE018
, according to the formula:
Figure 195537DEST_PATH_IMAGE019
Create a detection area for
Figure 95360DEST_PATH_IMAGE020
, the number of detected pedestrians is
Figure 349886DEST_PATH_IMAGE004
When , the corresponding Gaussian counting model is
Figure 385976DEST_PATH_IMAGE006
.6.根据权利要求1-3其中之一所述的人流量统计方法，其特征在于，用对应的高斯计数模型子集分析当前运动目标的人数，具体为：根据运动目标外接矩形框中心点坐标确定所处的子区域
Figure 12129DEST_PATH_IMAGE021
，在多高斯计数模型中获得子区域
Figure 336931DEST_PATH_IMAGE021
对应的高斯计数模型子集
Figure 441153DEST_PATH_IMAGE022
，即
Figure 784279DEST_PATH_IMAGE022
=(
Figure 581334DEST_PATH_IMAGE023
，
Figure 127852DEST_PATH_IMAGE024
，
Figure 301345DEST_PATH_IMAGE025
，
Figure 312026DEST_PATH_IMAGE026
),利用公式：6. The method for counting the flow of people according to any one of claims 1-3, wherein the number of people of the current moving target is analyzed with the corresponding Gaussian counting model subset, specifically: according to the coordinates of the center point of the circumscribed rectangular frame of the moving target Determine the sub-region you are in
Figure 12129DEST_PATH_IMAGE021
, to obtain subregions in a multi-Gaussian counting model
Figure 336931DEST_PATH_IMAGE021
The corresponding subset of Gaussian count models
Figure 441153DEST_PATH_IMAGE022
,Right now
Figure 784279DEST_PATH_IMAGE022
=(
Figure 581334DEST_PATH_IMAGE023
,
Figure 127852DEST_PATH_IMAGE024
,
Figure 301345DEST_PATH_IMAGE025
,
Figure 312026DEST_PATH_IMAGE026
), using the formula:
Figure 968398DEST_PATH_IMAGE027
   
Figure 64530DEST_PATH_IMAGE028
Figure 968398DEST_PATH_IMAGE027
   
Figure 64530DEST_PATH_IMAGE028
计算当前运动目标中包含的人数
Figure 448238DEST_PATH_IMAGE029
。Calculate the number of people included in the current exercise goal
Figure 448238DEST_PATH_IMAGE029
.7.根据权利要求1所述的人流量统计方法，其特征在于，快速目标跟踪关联方法，具体为：假设P1和P2分别为前后两帧图像中运动目标外接矩形的右上角顶点，其坐标分别为
Figure 579005DEST_PATH_IMAGE030
和
Figure 904813DEST_PATH_IMAGE031
，如满足条件：（1）前后两帧图像中的运动目标区域与检测线的交线段重叠或者部分重叠；（2）坐标
Figure 488241DEST_PATH_IMAGE032
，则为同一个目标。7. The method for counting the flow of people according to claim 1, characterized in that the fast target tracking association method is specifically as follows: Assume that P1 and P2 are respectively the upper right corner vertices of the moving target circumscribed rectangle in the two frames of images before and after, and their coordinates are respectively for
Figure 579005DEST_PATH_IMAGE030
and
Figure 904813DEST_PATH_IMAGE031
, if the conditions are met: (1) The moving target area in the two frames of images before and after overlaps or partially overlaps with the intersection line segment of the detection line; (2) the coordinates
Figure 488241DEST_PATH_IMAGE032
, then the same target.
CN201110423349.2A2011-12-162011-12-16Pedestrian volume statistical method based on plurality of Gaussian counting modelsActiveCN102542289B (en)

Priority Applications (1)

Application NumberPriority DateFiling DateTitle
CN201110423349.2ACN102542289B (en)2011-12-162011-12-16Pedestrian volume statistical method based on plurality of Gaussian counting models

Applications Claiming Priority (1)

Application NumberPriority DateFiling DateTitle
CN201110423349.2ACN102542289B (en)2011-12-162011-12-16Pedestrian volume statistical method based on plurality of Gaussian counting models

Publications (2)

Publication NumberPublication Date
CN102542289Atrue CN102542289A (en)2012-07-04
CN102542289B CN102542289B (en)2014-06-04

Family

ID=46349139

Family Applications (1)

Application NumberTitlePriority DateFiling Date
CN201110423349.2AActiveCN102542289B (en)2011-12-162011-12-16Pedestrian volume statistical method based on plurality of Gaussian counting models

Country Status (1)

CountryLink
CN (1)CN102542289B (en)

Cited By (38)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
CN103186903A (en)*2013-04-222013-07-03苏州科技学院Moving object detecting and tracking system based on PID (Proportion Integration Differentiation)
CN103400113A (en)*2013-07-102013-11-20重庆大学Method for detecting pedestrian on expressway or in tunnel based on image processing
CN103839035A (en)*2012-11-222014-06-04富士通株式会社Person number statistical method and person number statistical system
CN104123569A (en)*2014-07-212014-10-29中国人民解放军63908部队Video person number information statistics method based on supervised learning
CN104239905A (en)*2013-06-172014-12-24上海盖普电梯有限公司Moving target recognition method and intelligent elevator billing system having moving target recognition function
CN104318240A (en)*2014-07-182015-01-28天津理工大学Flower bud discriminating method based on computer vision
CN104506820A (en)*2015-01-062015-04-08成都新舟锐视科技有限公司Intelligent multi-target tracing system
CN104965964A (en)*2015-08-062015-10-07山东建筑大学Building personnel distribution model establishment method based on surveillance video analysis
CN105551266A (en)*2015-12-082016-05-04合肥寰景信息技术有限公司Method of calculating pedestrian flow threshold of traffic signal controller
CN105574499A (en)*2015-12-152016-05-11东华大学Method and system for detecting and counting number of people based on SOC
CN105912975A (en)*2015-02-202016-08-31空中客车集团印度私人有限公司Management of aircraft in-cabin activities occuring during turnaround using video analytics
CN106600576A (en)*2016-10-252017-04-26上海萨戈电子科技有限公司RGB-D camera-based human head locking method
CN106600643A (en)*2016-10-252017-04-26长安大学People counting method based on trajectory analysis
CN106600590A (en)*2016-12-122017-04-26四川华雁信息产业股份有限公司Automatic detection method and apparatus for oil leakage of current transformer
CN106845374A (en)*2017-01-062017-06-13清华大学Pedestrian detection method and detection means based on deep learning
CN106846297A (en)*2016-12-212017-06-13深圳市镭神智能系统有限公司Pedestrian's flow quantity detecting system and method based on laser radar
CN106845620A (en)*2016-12-192017-06-13江苏慧眼数据科技股份有限公司A kind of passenger flow counting method based on quene state analysis
CN107123126A (en)*2017-03-292017-09-01天棣网络科技(上海)有限公司A kind of stream of people's moving scene temperature method of estimation
CN107180420A (en)*2016-03-092017-09-19顺丰科技有限公司Article transmits householder method
CN107729804A (en)*2017-08-312018-02-23广东数相智能科技有限公司A kind of people flow rate statistical method and device based on garment ornament
CN107770921A (en)*2017-12-072018-03-06四川谋席科技有限公司A kind of energy-saving LED illumination system based on Cloud Server
CN107809834A (en)*2017-12-072018-03-16四川谋席科技有限公司A kind of resident illuminator based on cloud computing
WO2018059408A1 (en)*2016-09-292018-04-05北京市商汤科技开发有限公司Cross-line counting method, and neural network training method and apparatus, and electronic device
CN108229368A (en)*2017-12-282018-06-29浙江大华技术股份有限公司A kind of image display method and device
CN108549835A (en)*2018-03-082018-09-18深圳市深网视界科技有限公司Crowd counts and its method, terminal device and the storage medium of model construction
CN109389016A (en)*2017-08-102019-02-26株式会社日立制作所A kind of method and system that the number of people counts
CN109643321A (en)*2017-02-152019-04-16富士通株式会社 Video analysis system and video analysis method based on video surveillance
CN109753842A (en)*2017-11-012019-05-14深圳先进技术研究院 A method and device for counting people
CN109800683A (en)*2018-12-302019-05-24昆明物理研究所A kind of infrared pedestrian detection method and device based on FPGA
CN110008766A (en)*2019-03-262019-07-12南京大学 An Indoor People Counting Method Based on Radio Frequency Identification Technology
CN110826581A (en)*2018-08-102020-02-21京东数字科技控股有限公司Animal number identification method, device, medium and electronic equipment
CN111260628A (en)*2020-01-152020-06-09北京林业大学Large nursery stock number counting method based on video image and electronic equipment
CN111680569A (en)*2020-05-132020-09-18北京中广上洋科技股份有限公司Attendance rate detection method, device, equipment and storage medium based on image analysis
CN112005247A (en)*2020-06-122020-11-27深圳盈天下视觉科技有限公司People flow data monitoring system and people flow data display method and device thereof
CN113536891A (en)*2021-05-102021-10-22新疆爱华盈通信息技术有限公司Pedestrian traffic statistical method, storage medium and electronic equipment
CN113902772A (en)*2021-11-192022-01-07深圳市联洲国际技术有限公司Crowd counting method and device, computer storage medium and processor
WO2022042352A1 (en)*2020-08-282022-03-03安翰科技(武汉)股份有限公司Image recognition method, electronic device and readable storage medium
CN114943930A (en)*2022-05-052022-08-26北京埃福瑞科技有限公司Station passenger clearing method and device

Citations (3)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
US6674877B1 (en)*2000-02-032004-01-06Microsoft CorporationSystem and method for visually tracking occluded objects in real time
CN101303727A (en)*2008-07-082008-11-12北京中星微电子有限公司Intelligent management method based on video human number Stat. and system thereof
CN101325690A (en)*2007-06-122008-12-17上海正电科技发展有限公司Method and system for detecting human flow analysis and crowd accumulation process of monitoring video flow

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
US6674877B1 (en)*2000-02-032004-01-06Microsoft CorporationSystem and method for visually tracking occluded objects in real time
CN101325690A (en)*2007-06-122008-12-17上海正电科技发展有限公司Method and system for detecting human flow analysis and crowd accumulation process of monitoring video flow
CN101303727A (en)*2008-07-082008-11-12北京中星微电子有限公司Intelligent management method based on video human number Stat. and system thereof

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
高陈强等: "视频图像中基于特征匹配的人流量统计", 《电视技术》*

Cited By (51)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
CN103839035A (en)*2012-11-222014-06-04富士通株式会社Person number statistical method and person number statistical system
CN103186903A (en)*2013-04-222013-07-03苏州科技学院Moving object detecting and tracking system based on PID (Proportion Integration Differentiation)
CN104239905A (en)*2013-06-172014-12-24上海盖普电梯有限公司Moving target recognition method and intelligent elevator billing system having moving target recognition function
CN103400113B (en)*2013-07-102016-08-24重庆大学Freeway tunnel pedestrian detection method based on image procossing
CN103400113A (en)*2013-07-102013-11-20重庆大学Method for detecting pedestrian on expressway or in tunnel based on image processing
CN104318240A (en)*2014-07-182015-01-28天津理工大学Flower bud discriminating method based on computer vision
CN104123569A (en)*2014-07-212014-10-29中国人民解放军63908部队Video person number information statistics method based on supervised learning
CN104506820A (en)*2015-01-062015-04-08成都新舟锐视科技有限公司Intelligent multi-target tracing system
CN105912975B (en)*2015-02-202021-10-26空中客车集团印度私人有限公司Managing aircraft in-cabin behavior occurring during a station crossing using video analysis
CN105912975A (en)*2015-02-202016-08-31空中客车集团印度私人有限公司Management of aircraft in-cabin activities occuring during turnaround using video analytics
CN104965964A (en)*2015-08-062015-10-07山东建筑大学Building personnel distribution model establishment method based on surveillance video analysis
CN104965964B (en)*2015-08-062018-01-30山东建筑大学A kind of building personnel's distributed model method for building up based on monitor video analysis
CN105551266A (en)*2015-12-082016-05-04合肥寰景信息技术有限公司Method of calculating pedestrian flow threshold of traffic signal controller
CN105574499A (en)*2015-12-152016-05-11东华大学Method and system for detecting and counting number of people based on SOC
CN105574499B (en)*2015-12-152019-08-20东华大学 A method and system for detecting and counting people based on SOC
CN107180420A (en)*2016-03-092017-09-19顺丰科技有限公司Article transmits householder method
WO2018059408A1 (en)*2016-09-292018-04-05北京市商汤科技开发有限公司Cross-line counting method, and neural network training method and apparatus, and electronic device
CN106600643A (en)*2016-10-252017-04-26长安大学People counting method based on trajectory analysis
CN106600576A (en)*2016-10-252017-04-26上海萨戈电子科技有限公司RGB-D camera-based human head locking method
CN106600590B (en)*2016-12-122019-09-06四川华雁信息产业股份有限公司 Current transformer oil leakage automatic detection method and device
CN106600590A (en)*2016-12-122017-04-26四川华雁信息产业股份有限公司Automatic detection method and apparatus for oil leakage of current transformer
CN106845620A (en)*2016-12-192017-06-13江苏慧眼数据科技股份有限公司A kind of passenger flow counting method based on quene state analysis
CN106845620B (en)*2016-12-192019-09-10江苏慧眼数据科技股份有限公司A kind of passenger flow counting method based on quene state analysis
CN106846297A (en)*2016-12-212017-06-13深圳市镭神智能系统有限公司Pedestrian's flow quantity detecting system and method based on laser radar
CN106845374A (en)*2017-01-062017-06-13清华大学Pedestrian detection method and detection means based on deep learning
CN106845374B (en)*2017-01-062020-03-27清华大学Pedestrian detection method and detection device based on deep learning
CN109643321A (en)*2017-02-152019-04-16富士通株式会社 Video analysis system and video analysis method based on video surveillance
CN107123126A (en)*2017-03-292017-09-01天棣网络科技(上海)有限公司A kind of stream of people's moving scene temperature method of estimation
CN109389016B (en)*2017-08-102023-04-07株式会社日立制作所Method and system for counting human heads
CN109389016A (en)*2017-08-102019-02-26株式会社日立制作所A kind of method and system that the number of people counts
CN107729804A (en)*2017-08-312018-02-23广东数相智能科技有限公司A kind of people flow rate statistical method and device based on garment ornament
CN109753842A (en)*2017-11-012019-05-14深圳先进技术研究院 A method and device for counting people
CN107809834A (en)*2017-12-072018-03-16四川谋席科技有限公司A kind of resident illuminator based on cloud computing
CN107770921A (en)*2017-12-072018-03-06四川谋席科技有限公司A kind of energy-saving LED illumination system based on Cloud Server
CN108229368B (en)*2017-12-282020-05-26浙江大华技术股份有限公司Video display method and device
CN108229368A (en)*2017-12-282018-06-29浙江大华技术股份有限公司A kind of image display method and device
CN108549835A (en)*2018-03-082018-09-18深圳市深网视界科技有限公司Crowd counts and its method, terminal device and the storage medium of model construction
CN110826581A (en)*2018-08-102020-02-21京东数字科技控股有限公司Animal number identification method, device, medium and electronic equipment
CN110826581B (en)*2018-08-102023-11-07京东科技控股股份有限公司Animal number identification method, device, medium and electronic equipment
CN109800683A (en)*2018-12-302019-05-24昆明物理研究所A kind of infrared pedestrian detection method and device based on FPGA
CN110008766B (en)*2019-03-262020-04-24南京大学Indoor people counting method based on radio frequency identification technology
CN110008766A (en)*2019-03-262019-07-12南京大学 An Indoor People Counting Method Based on Radio Frequency Identification Technology
CN111260628A (en)*2020-01-152020-06-09北京林业大学Large nursery stock number counting method based on video image and electronic equipment
CN111680569A (en)*2020-05-132020-09-18北京中广上洋科技股份有限公司Attendance rate detection method, device, equipment and storage medium based on image analysis
CN111680569B (en)*2020-05-132024-04-19北京中广上洋科技股份有限公司Attendance rate detection method, device, equipment and storage medium based on image analysis
CN112005247A (en)*2020-06-122020-11-27深圳盈天下视觉科技有限公司People flow data monitoring system and people flow data display method and device thereof
WO2022042352A1 (en)*2020-08-282022-03-03安翰科技(武汉)股份有限公司Image recognition method, electronic device and readable storage medium
CN113536891B (en)*2021-05-102023-01-03新疆爱华盈通信息技术有限公司Pedestrian traffic statistical method, storage medium and electronic equipment
CN113536891A (en)*2021-05-102021-10-22新疆爱华盈通信息技术有限公司Pedestrian traffic statistical method, storage medium and electronic equipment
CN113902772A (en)*2021-11-192022-01-07深圳市联洲国际技术有限公司Crowd counting method and device, computer storage medium and processor
CN114943930A (en)*2022-05-052022-08-26北京埃福瑞科技有限公司Station passenger clearing method and device

Also Published As

Publication numberPublication date
CN102542289B (en)2014-06-04

Similar Documents

PublicationPublication DateTitle
CN102542289A (en)Pedestrian volume statistical method based on plurality of Gaussian counting models
CN102214309B (en)Special human body recognition method based on head and shoulder model
CN107909081B (en) A fast acquisition and fast calibration method for image datasets in deep learning
CN105404847B (en)A kind of residue real-time detection method
CN109145708B (en)Pedestrian flow statistical method based on RGB and D information fusion
JP2019505866A (en) Passerby head identification method and system
CN104978567B (en)Vehicle checking method based on scene classification
CN103268470B (en)Object video real-time statistical method based on any scene
CN113139521A (en)Pedestrian boundary crossing monitoring method for electric power monitoring
CN104637058B (en)A kind of volume of the flow of passengers identify statistical methods based on image information
CN107169985A (en)A kind of moving target detecting method based on symmetrical inter-frame difference and context update
CN104835147A (en)Method for detecting crowded people flow in real time based on three-dimensional depth map data
CN108985170A (en)Transmission line of electricity hanger recognition methods based on Three image difference and deep learning
CN106991370B (en)Pedestrian retrieval method based on color and depth
CN103310194A (en)Method for detecting head and shoulders of pedestrian in video based on overhead pixel gradient direction
Xu et al.Real-time pedestrian detection based on edge factor and Histogram of Oriented Gradient
CN103310444A (en)Method of monitoring pedestrians and counting based on overhead camera
CN102609720A (en)Pedestrian detection method based on position correction model
CN111028263B (en)Moving object segmentation method and system based on optical flow color clustering
CN102194270B (en)Statistical method for pedestrian flow based on heuristic information
Yang et al.A method of pedestrians counting based on deep learning
CN111626107B (en)Humanoid contour analysis and extraction method oriented to smart home scene
CN104200483B (en)Object detection method based on human body center line in multi-cam environment
Lee et al.Real-time automatic vehicle management system using vehicle tracking and car plate number identification
CN110378935B (en)Parabolic identification method based on image semantic information

Legal Events

DateCodeTitleDescription
C06Publication
PB01Publication
C10Entry into substantive examination
SE01Entry into force of request for substantive examination
C14Grant of patent or utility model
GR01Patent grant
[8]ページ先頭
©2009-2025 Movatter.jp