Research On Extraction Technology Of Moving Objects In The Video Synopsis

With the development of intelligent surveillance, it has been more widely used so that vast amounts of surveillance video were produced every day. Confronted with huge numbers of video data, it is extremely difficult to browse and investigate entirely manually. People began to use different techniques to reduce the manpower and time cost in the process of browsing and identifying. Video synopsis is one of the most effective methods, in which moving object detection and motion trajectory extraction are the most critical steps. The ability to extract moving object effectively from continuous complete motion frame sequences will directly affect the quality of the video synopsis. Meanwhile, moving object detection and extraction, as the foundation and the key technology in the field of intelligent surveillance and security, can effectively promote the scene understanding, behavior analysis, criminal investigation, etc. Thus, it has been of great value and research significance in the field of security, transportation, military, etc.Nowadays, complete motion sequences sometimes can’t be extracted accurately and effectively due to the presence of ambient occlusion, shape changes, illumination variation and other interference factors in the complex surveillance scene. A motion process may be extracted into two or more different motion frame sequences so that a sudden appearance or a sudden disappearance may occur in the generated video synopsis which are seriously inconsistent with the motion process in the original video and can cause some interference and confusion in motion analysis and scene understanding. To solve this problem, we conduct a depth research and analysis, of which the main work and innovation are as follows:1. Particle filter tracking algorithm is improved by combining with the ViBe and compress tracking. Foreground result detected by ViBe is used to guide particles’transfer and sampling, which can effectively solve the problem of relocation of objects after the shelter; And with the feature sampling based on compressed sensing can capture multi-scale changes features of the object, which can effectively characterize the appearance model of the moving object and improve the tracking accuracy. Meanwhile, combining foreground result with tracking can effectively handle the movement changes in the tracking process, which can track the target object stably and accurately in the scene under occlusion.2. A novel tracking and extracting method based on the combination of Vibe algorithm and improved particle filter tracking is proposed. With the assistance and feedback of the moving object’s foreground information and status information in tracing process, we conduct the final extraction and classification of the motion frame sequences based on the foreground extraction and motion tracking results, which can solve the occlusion problem effectively in the process of extraction, and ensure the integrity and continuity of the moving object’s motion frame sequences.Experiments show that the algorithm can extract the complete continuous motion frame sequences of moving object under both partial occlusion and global shelter conditions in complex scene with good performances.