Research On Video Synopsis System Based On Color Features Of Moving Objects

With the development of computer vision and the attention of national government departments to the cause of security.In response to the relevant government policies,intelligent surveillance system has been widely used and studied,while a large number of cameras are installed in every corner of our lives.The explosive growth of video surveillance data brings great challenges to the follow-up processing.Video Synopsis technology processes the original video,rearranges and combines the interested objects to form a highly concentrated video.It provides a new idea to solve the difficulties of storage,retrieval and transportation of massive video surveillance data,and becomes an important part of intelligent surveillance system.This paper mainly expounds the principle of three core technologies of Video Synopsis system,and focuses on the key steps of moving object detection,moving object tracking and trajectory combination in the whole video concentrating process.Aiming at the shortcomings of video concentrating technology,the algorithm of video concentrating is deeply studied and optimized to complete the construction of Video Synopsis system.The main work of this paper is as follows:?As the first step in Video Synopsis technology,moving object detection is also the foundation of building Video Synopsis system.Its purpose is to extract the complete background and moving target,and to distinguish them.This paper chooses the moving object detection technology based on the fusion of CLD and ViBe algorithm to compensate for the interference caused by the ghost brought by the original ViBe algorithm.This method extracts video key frames by using CLD color space descriptor,selects key frames for three frame difference,obtains the real background model,and then detects the moving target with ViBe.In the updating stage of the background model,morphological processing technology of adaptive threshold is added to update the background model in real time and eliminate the interference information in the dynamic background model.?Because the particle filter algorithm with single feature can not adapt to the change of direction and scale of moving object in video scene,all the particle filter tracking algorithm with color feature of moving object is used to track moving object.Under the theoretical framework of particle filter tracking algorithm with single feature description,two kinds of particle filters are established for moving targets,namely HOG particle filter and HSV particle filter.The Barkholdian distance of the two eigenvalues is calculated,the similarity comparison and weight calculation are carried out,and the final tracking results are obtained,which improves the robustness of the tracking algorithm and reduces the missing rate.? In the aspect of trajectory combination optimization,the trajectory combination optimization method based on energy cost function is adopted,and the optimal combination of trajectory combination is obtained by solving the minimum energy function.?According to the above results,Poisson fusion method is used to stitch the background image extracted from moving object detection and the moving object which obtains the optimal combination of trajectories together in accordance with the law to form a new concentrated video.