The Study For Surveillance Video Synopsis

With the development of technology and the safety awareness enhancement of the society, millions of surveillance cameras are widely used in various industries, and simultaneously capture millions of video with the full-time working. Due to the limitation of labour resource and original weakness of human’s physiology, It’s very difficult to find the key information in these massive videos. In order to browse these video data fast and rapidly has been the urgent requirement in the field of surveillance industry, and it is also very meaningful to do the research about surveillance video synopsis.The main work in this thesis is to introduce the current researches about video synopsis, and combine the particular characteristic of surveillance video, we propose the framework of surveillance video synopsis based on the moving object. Firstly, the background image is produced by using the Gaussian mixture model; and then we detect the moving objects with the background subtraction; and then track the moving objects using the combining with Kalman filtering of MeanShift algorithms, and the activity object is matted by grabcut method. As the surveillance background is fixed, the extracted moving objects are shifted on the time axis, it can make multiple objects be appeared in the same frame simultaneously. At last, the frame image is generated by stitching the objects’ image and the background image, which furtherly make up the video synopsis.After the realization of the entire surveillance video synopsis, the obtained synopsis showed it can retain the key information and reduce the redundancy as far as possible.