Research And Implementation Of Video Synopsis System In Intelligent Video Surveillance

With Intelligent Video Surveillance System(IVSS) becoming increasingly popular in the security work, checking and retrieving vast amounts of video data is more and more time-consuming and laborious, and artificial work makes mistakes such as flaws and delays frequently. The appearance of Video Synopsis Technology offers an important way to solve this problem. In recent years, a lot of findings based on Video Synopsis have been researched, but still have certain gap for the requirement of actual monitoring application.This thesis firstly expounds the basic principle of Video Synopsis, and introduces the overall architecture of the system systematically.Then we study several classical moving object detection and tracking algorithms. At last we worked out that the background modeling method based on random sampling(Vi Be) is more advantageous by analyzing the detection rate and the false alarm rate in different monitoring scenes; By analyzing the tracking accuracy rate and the success rate, we worked out that the method combining the Mean Shift and Particle Filter based on correlation is more conducive to extract complete trajectory of moving target in monitor scene.Thirdly a method on trajectory restructuring which can guarantee targets chronological and an optimizing method for fusion between targets and background are proposed. We rearrange the target's occurrence time in synopsis video by controlling the degree of density and overlap rate based on keeping chronological, and then we analyze the performance between the trajectory restructuring method based on the energy function and the proposed approach with experimental comparison. The experiment shows that the proposed approach can be chronological and more efficient. We combine Poisson Image Editing with the weighted fusion algorithm by judging the state of pixel, aiming to reduce seams between target and background and weaken shades between targets, we compare and analyze the performance between Poisson Image Editing not combined with the method of weighted fusion and the proposed method, the research shows that the proposed approach can achieve better results under the condition that there are some shades in the final synopsis video and that there are a large number of differences on luminance between target and background.Finally we introduce the software interface and the main function of the Video Synopsis System that we build which have some kinds of functions including feature retrieval, snapshot and playback, and briefly expound the basic principle of each function module, then assess the effectiveness of each module by taking an example of color extracting.Experiments demonstrate that our Video Synopsis System can generate clearer synopsis video, and the accuracy of its function like feature retrieval and snapshot is higher, which make it meet the requirements of monitoring applications better.