The People Counting System Based On Video Image

In recent years,With the accelerating of urban informatization,the system has become the basis of data analysis and process for public transport services,commercial passenger flow analysis,public safety protection and other industries.The statistics of people based on video images is an important topic in modern intelligence research.It has the advantages of high stability,high precision and low cost.However,in practical application,due to individual differences and occlusion between pedestrians,the pedestrian target will be interfered in the detection,recognition and tracking stage,which will ultimately affect the accuracy of statistical results.At the same time,how to balance the real-time and accuracy in the algorithm processing module is still the key problem to be solved in the system construction.In this paper,video pedestrian detection based on region of interest(ROI)is completed by training classifier of head and shoulders feature extraction,and combined with the tracking based on feature matching to calculate the number statistics.Finally,a people counting system based on video image is implemented.Its main research work is as follows:1.In pedestrian detection: Firstly,build the pedestrian head shoulders sample database,the histogram of oriented gradient(HOG)feature extraction method is used to detect the fixed head shoulder regions of the human body(like the "?" shape)and train the support vector machine to obtain the pedestrian head shoulder detector.This method can effectively reduce the influence of shape change and occlusion.Then,three-frame difference combined with background difference algorithm is used to obtain the foreground target region.Preliminary reduction of detection range.Finally,the head shoulder detector obtained previously is used for centralized detection of the marked ROI rectangular.This method can reduce the time consumption of the algorithm based on video pedestrian detection.2.In people counting: A tracking statistics method based on pedestrian detection feature matching is proposed.Firstly,feature matching was conducted on outline area and distance of centroid point,the mismatching state caused by occlusion was classified and judged,and the trajectory was drawn to complete the tracking.Then,the number of people in different status is calculated,and the number is updated after passing through the ROI area.Combined with the tracking algorithm for people counting.the error detection results of pedestrian detection can be corrected and it basically satisfies the counting requirement under partial occlusion interference.3.In the development environment of Visual Studio 2013 and OpenCV,the predefined MFC class is used to design the people counting system based on video image,and has better accuracy and high efficiency.