Research On Moving Object Retrieval For Video Surveillance

Intelligent video surveillance system is a research and application field of computer vision that attract much attention recently, which is widely applied in military and civil place and has a broad application prospects and profound theoretical value. It's very important for public safety to select existing target from the surveillance system. So that for the target retrieval problems in video surveillance system, we propose the moving targets retrieval technologies that based on moving object detection, target tracking and object matching. The main contents are summarized as follows:It's very important to public safety is of special significance that select existing target from the surveillance system, so in video monitoring system of target retrieval problems, so we put forward by moving object detection, target tracking, object matching to complete moving targets retrieval technologies. The main contents are summarized as follows:First of all, we study target detection technology for both static and moving background under video surveillance. Through the realization of different algorithm and the analysis of results, and for the final Requirements, we propose the detection technology based on gaussian mixture model for extracting of moving object. This algorithm can effectively handle snow background, water fluctuation, branches swaying scene by online learning and training process, and got robust performance compared with other algorithms. This is the target center and tracking template for the future tracking.Secondly, we propose adaptive meanshift tracking algorithm based on template update. By analyze the results of SIFT points matching, we establish the least-square module and we update target template by using linear fitting method for current template and detection results. This step completed the confirmation of the same target confirmation and the elimination of false targets, and provides possible target template for the subsequent object matching.Finally, we studied target matching technology based on grey image matching and features image matching strategy, and realized the algorithm combined with the requirements. This paper studied the preliminary retrieval technology, and for complex scenes such as people wearing similar clothes, or existing interaction etc, we need further study. The result of detection, tracking and matching are interdependence, how to use the information to constitute effective feedback loop that help improve retrieval precision.