Extraction Method Of Motion Trajectory In Surveillance Video

Motion trajectory extraction plays a very important role in video surveillance technology, which involves moving target detection, target classification, target matching, etc. In this thesis, the methods for extracting motion trajectories of single and multiple human targets for surveillance are researched and complicated problems in multi-target occluded scene are successfully solved.In trajectory extraction, it is necessary to locate each target in each frame. Given the complicacy of moving targets in the surveillance scene, a trajectory extraction model based on multi-tuple is proposed in this thesis. In this model, not only is the position information recorded, but also the information of motion direction, speed, external rectangle, contour, and of whether the target is occluded is included. Therefore, the definition of trajectory is enriched in this model and it is conducive to further analysis.A novel Camshift method based on spatiotemporal constraint strategy is proposed so as to solve the window-shift problem of the classic target tracking method-Camshift. In this method, the searching area is restricted by anticipating the possible position of target, and the results are more close to reality by restricting the information of size according to the results of motion detection. The results of simulate experiment demonstrate that both speed and precision are greatly improved.Moreover, a trajectory extraction method of multiple targets is presented to solve the complicated problems of low precision in occluded multi-target motion trajectory extraction. The tracking modes can be adaptively shifted according to whether the target is obstructed or not. The results of simulate experiment show that the motion trajectory can be robustly extracted even though multiple human targets are occluded.