Pedestrian Tracking Via Multi-camera In Video Surveillance

Video surveillance is an essential part of the social security. With the rapiddevelopment of social economy and the increasing range of video surveillance, smartmonitor becomes an urgent need. Especially, multi-camera video surveillance for the samelocation is very necessary in some important places. Therefore, pedestrian tracking viamulti-camera has become an important issue in computer vision.Multi-camera tracking is often based on feature matching or three-dimensionalreconstruction method. As information fusion from multi-camera by feature matching,when the angle difference between cameras becomes large, objects’ characteristics changea lot, the matching error rate will increase. As information fusion by three-dimensionalreconstruction, it has lower error rate than the former on, but has high computationalcomplexity and needs additional information. For these reasons, a pedestrian tracking viaMulti-camera method based on homography constraint has been realized.First, the object information is needed for object tracking. By experiments, itcompared variant pedestrian object extraction method and its difference when theparameters changed. Second, it calculated the homography matrix between cameras, andused homography constraint rules to fuse information from multi-camera and got a footmap. Then, it did the reverse perspective projection to get the corresponding relationshipbetween the objects and the foot map. To improve the accuracy of the correspondingrelationship, a correction strategy is proposed. It defined different confidence of trackingresult from different cameras, and finally tracking objects with multi camerassynergistically. Experiment shows that its time complexity is not high. Especially, it cantrack the object accurately when the target block is occluded in some cameras.