Research Of Surveillance Video Synopsis Algorithm Based On Object

This paper presents a surveillance video synopsis algorithm based on object on the basis of the traditional video abstract algorithm.In our algorithm,the original video frame is not considered as the smallest unit,the original frame has been changed,and the motion objects have been recombination,the length of video has been further compressed.At first,the foreground image of moving targets is segmented,and all the foreground targets are tracked.Then the segmented foreground images and its trajectory are saved.When the number of saved objects is larger than the given threshold of synopsis density,all the saved foreground images are attached to the appropriate background according to their corresponding trajectories.So,a dynamic synopsis video is formed.The synopsis video includes more dense moving targets,the video duration is shorter and retains the original dynamic information.This thesis mainly includes the following aspects:1)The classic foreground segmentation algorithms have poor results in the low contrast of foreground and background and the dark environment.The new foreground segmentation algorithm of combining color features and texture features of images adopted in our thesis,the accurate segmentation results will be obtained in the low contrast environment.2)The classic MeanShift algorithm,Kalman Filter algorithm and Particle Filter algorithm are analyzed in detail.On the basis of the advantages and disadvantages and the adaptation in anti-occlusion tracking of these classic tracking algorithm,a collision analysis algorithm based on moving block which combines Kalman Filter with Particle Filter is presented.Our algorithm can select self-adaptively the different tracking algorithm when detecting whether there is occlusion.The experimental results show that the new algorithm achieved the anti-occlusion tracking and enhanced the robustness of tracking.3)In our video synopsis algorithm,the foreground images,foreground mask and motion trajectory are all saved in process of multi-target tracking.When the number of saved objects reaches the previously set threshold of synopsis density,all the foreground images of the early preservation are joined together to the appropriate background image according to the corresponding their trajectories,so a dynamic synopsis video is formed.This synopsis video has greatly increased the utilization in time and space of the moving target,further shorten the video duration,and more convenient for people to browsing and retrieval the key information in surveillance video.4)An intelligent surveillance video analysis and synopsis system is developed by integrating all of algorithms above based on OpenCV and MFC desktop development technology.This system software has the functions of moving foreground segmenting,multi-targets tracking,moving trajectory generating,moving objects counting,video synopsis forming,and provides a good interface for developing the more powerful video retrieval system.