Human Flow Statistic And Recognition System Based On Binocular Vision

Along with the rapid development of economy, there are more and more large-scale activities take place in the city, and business competition becomes more and more incandescent. So the human flow statistic system which can offer the accurate human flow information to the security and business plan departments becomes hotspot of the research area. Simultaneity, binocular vision also becoming a research hotspot of the computer vision domain, and it also be widely applied in many domains. Therefore the research of this subject has the important theory significance and the practical value.First, this paper focused on stereo matching algorithm which is the most basic issue in stereo vision. The theory and four key components of stereo matching algorithm, including feature space, similarity measure, search space and search strategy are introduced in detail. On basis of these key components, this paper divided the stereo matching algorithms in common use into three main categories, that is, intensity-based matching, feature-based matching and domain transform-based matching. And the basic idea, characteristics, advantages and disadvantages of each kind of algorithm is explained in detail. Second, on basis of abundant inland and overseas reference papers, this paper introduced the most difficult problem - calibration in two chapters. They introduced how to get the inner and outer parameters of camera respectively which called camera calibration. Then we can calculate the 3D position of any feature point in the 2D image and the problem of target objects partial coverage of each other is solved. Third, this paper introduced the theory and algorithm of Mean Shift. It also adopts a pattern clustering technique based on Mean Shift successfully. At last, our system is proved to be effective and robust based by doing experiments with large numbers of sample images we captured.