Video Synopsis Technology Based On Moving Object

With the concept of "Green City" is proposed, ten of thousands of monitoring equipments are installed in banks, schools, corridors, so it will produce a lot of surveillance videos every day, which gives us a great challenge in video handling works such as the browsing, retrieval and storage. Video synopsis is an effective technology that can solve these problems, which can greatly shorten the length of the surveillance video, meanwhile, it may not lose the motion informations of original video. The technology not only achieve effective storage, but also be easier for users to browse and retrieve the surveillance video, which has become a hot technology of handing the surveillance video now.This paper firstly elaborates the synopsis principle of surveillance video, and then introduces several key steps of the video synopsis, such as moving object detection, moving object tracking, combination of the trajectory optimization. Aim at this limitations of the existing algorithms in every step, this paper proposes the improved algorithms correspondly and gets the the final synopsisi result. The main work is as follows:Since the background modeling method based on single feature has some limitations, so in order to improve the robustness and accuracy of moving object detection, this paper presents a background modeling algorithm which combinates the texture feature of SI LTP and the color feature of single Gauss. In this method, it will firstly extract the moving foreground and background using the texture model of SILTP, and then use the color model of single Gauss to detect the targets when the target texture is similar to the background texture, so it will correct the misjudged foreground and background. The experimental results show that: comparing with other background modeling method of single feature, the algorithm of this paper has better effect and the the contours of the moving objects detection is more clear, the recall rate and the false alarm rate are better than the background model based on single feature.In order to improve the accuracy of racking moving objects, this paper proposes a trajectory tracking algorithm which based on the linear fusion of Camshift and SIFT. In this algorithm, Firstly, Camshift algorithm is used to track the moving target preliminary and get the tracking area. Secondly, using the SIFT feature vector to match the target area and tracking area, so the SIFT matching and correction results are obtained. Lastly, the target tracking results are linear fusion of the two algorithm, which get the ultimate results. The experimental results show that: comparing with MeanShift and Camshift, the racking error of this algorithm is minimal and it can achieve the better tracking.In order to maintain the space consistency and ruduce collisions in synopsis video, this paper proposes a trajectory optimization method which based on the time axis. In this method, it will use the energy cost function to meature the conversion of original video to the synopsis video.When the energy cost function is minimal, the effect of trajectory optimization is best. Finally, update and extract the background image in the original video, then stitch the objects in the corresponding background image by the Poisson editing, so it will form the synopsis video. The experimental show that: comparing with the trajectory optimization method which based on temporal-spatial transfer, this method can effectively avoid the collision between targets and keep the revelance of the original video and make the video more real.