Research And Application On Object Collaborative Tracking For Surveillance Video

The intelligent video surveillance technology can automatically analyze the surveillance video data, recognize the interested objects and events, and make the object-pertinent storage and classification for convenient further search. Compared with the conventional video surveillance technology mainly depending on manual observation, it greatly improves the efficiency and accuracy of video surveillance, and can satisfy the gradually increased requirements in public security. It has a wide application prospect and great research significance, has attracted the attention of numerous foreign and domestic researchers. Moving objects tracking is considered as the basic problem and important content of intelligent video surveillance technology. Aiming at the problem of limited vision of single camera, this thesis introduces an object-collaborative object tracking method with multi-cameras. The main work consists of the following three parts:(1) For the problem that the disturbance of background pixels and the scale of the moving objects continuously changing during tracking, a posterior probability model based adaptive tracking algorithm is described. Combined with statistic information of object search region, the posterior probability of statistic feature can effectively reduce the disturbance of background pixels in object region during tracking. Then a new object tracking method with adaptive tracking window is proposed based on the posterior probability model of statistic feature. It uses specific strategies to adjust tracking windows according to scale change of the objects in different directions and makes the tracking more robust. Experiments have been conducted to verify the described algorithm at last.(2) For the surveillance scenes of which the multiple cameras have overlapped views, a candidate object set based multi-camera collaborative tracking method is proposed. For the problem of objects collaborative tracking, the traditional plane projective transformation based method is firstly introduced, followed by the analysis of the limited application and the homographic constraint conditions. Then the candidate object set based cameras collaborative tracking method is presented: First of all, the moving objects in the view of candidate camera are extracted by moving objects detection, and then the candidate objects are filtrated from the moving objects set according to the epipolar geometric constraints. At last, according to the posterior probability model of the statistic features, by searching in the candidate set of targets, the optimal matching result of tracking is obtained, the task of multi-camera collaborative tracking has completed. The experiments verify the effectiveness of proposed method.(3) A prototype system of collaborative object tracking is designed and implemented according to the proposed methods, followed by the analysis of its structure and workflow. This prototype system is deployed in the real scene. The performance of the prototype system in actual surveillance proves the practicality of proposed methods.