Video Synopsis Research Based On Object Motion Information

Facing the huge video data generated by the surveillance cameras that operate 24 hours a day, how to express the information compactly for the convenience of consequent browsing, back-check and analyzing, is becoming a hot and hard spot of research in the field of video surveillance. After many years of development, the traditional technology of video abstract can select key-frames or interesting video clips from the original video. But to generate a summary video with the maximized information, minimized redundancy and keeping the dynamic character, is still a great challenge.This work concentrates on the research of video synopsis based on object motion information, and the contributions are as follows:1. Summarized and analyzed current dominant video abstraction approaches, including key-frame based video summary, interesting video clips based video skimming, video montage and the original method of video synopsis based on object motion information.2. Proposed a set theoretical method for object start-time programming. The problem of visual information maximization is formulated in terms of set theory, and then translated into a discrete combinatorial optimization problem by using set operations. Finally, the problem is expressed as a fixed-point iteration by continuous relaxation. The proposed algorithm can converge to the local optimum within a few iterations. Comparative experiments are conducted, and appropriate evaluation function is defined to measure the performance of different approaches. Experimental results illustrate the validity of our method.3. Constructed the framework for the video synopsis based on object motion information, and finished the software coding, including the object and background extraction based on mixture Gaussian model, object start-time programming based on the set theory, background image resampling with respect to the time lapse and the real distribution of the object appearance time in the original video, image stitching between the object and background, and then got the synopsis video.