Investigation Of Moving Object Tracking In Multi-Scene Based On Multi-Camera

With the development of the discipline of computer vision, pattern recognition, image processing, computer vision has become a popular field of research in information science today, and visual tracking is a core in the computer vision field.In most cases, due to the limitation of single camera vision, we can neither observed the region of interest from a single camera nor tracked the moving target continuously.And it is also difficult to slove the occlusion of targets.Therefore, target tracking technology based on multi-camera has very important significance not only in academic research, but also in engineering application.Within the freamwork of video surveillance system, this thesis extended the multi-camera tracking, and it can be described into three key technologies:target detection, target location and object tracking.The main research works of this thesis can be described as follows:(1)Researched the technologies of object detection, and analysis some common targets detection algorithm.Proposed a method based on Gaussian mixture codebook background modeling which combines the advantages of the Gaussian mixture and the codebook.Experiments show that the method can accurately extract foreground information.(2)Analyzed the targets tracking method.To slove the tracking algorithm via spatio-temporal context can only achieve tracking problems of single target,we expanded the track method to multiple target tracking. The experiment shows that the method can meet the requirements of real-time target tracking.(3)Introduced the homography matrix which exchanged coordinate among multi-views.Due to the different of the monitored scene, we proposed an algorithm for detecting the calibration points automatically by using SIFT combined with manual selection, which improved the accuracy of homography matrix. In view of the fusion process, according to the reference view selection, presented a reference view selection method, which improved the accuracy of target localization.(4) Builded a multi-camera monitoring experimental platform, through the layered method, which supported the expansion of hardware devices. According to the experimental platform, we proposed a target synergy strategy so as to reduce the amount of calculation to a certain extent, saves the computing resources.