Study Of People-counting Algorithms Based-on Video Image Processing

With the improvement of digital image processing and pattern recognition, videoimage analysis has been widely and deeply studied. People counting is one importantarea in the domain of it.A people counting system includes a series of processes: Video Collection, PeopleDetection, People Tracking and Counting. The main contributions of this thesis aresummarized as follows.1) How to place the camera is important to the algorithm design. Final wedetermine to place the camera vertically with a small deflection. The camera placedvertically can reduce the loss of information effectively. The separation algorithm basedon human form needs the camera placed with a small deflection.2) If there are no effective separations of human body with mutual contact, thetracking algorithm is impossible to be accurate. This paper proposes an algorithm basedon the human morphology. It has used the two-dimensional characteristics of the humanbody to find out that the maximum pit just is the two-body contact points3) In the study of moving object tracking algorithm, the paper proposes acombination method of feature matching and region matching. This paper presents aglobal optimal matching method based on bipartite graph for matching the same objectin series images. To find the global optimal match, we can calculate the weightedaverage of characteristic variables and regional variables.4) In the system implement stage, this article uses OpenCV which is IntelCompany’s computer visual function and image processing library. The pthread packagein the Linux system is used to achieve a multithreading system. As a complete system,besides the core algorithm of pedestrians counting, it also needs other auxiliaryfunctions, including camera configuration, parameter configuration, results saving, andresults showing.