Research On Key Algorithms In Intelligent Video Surveillance

Intelligent video surveillance (IVS) is a highly intelligent system, which improve the efficiency of video surveillance and reduce the excessive use of human resources mainly. In order to make intelligent video surveillance system plays a good performance in all scenarios, which need the very powerful algorithms to support. The key algorithms ’of intelligent video surveillance system including moving target detecting algorithms, moving target tracking algorithms, moving target behavior analysis algorithms and graphics pre-processing algorithms. These algorithms determine intelligent video surveillance system’s performances and results. This thesis does the corresponding research on moving target detecting algorithm and video summary technology, and puts forward the solutions.In terms of moving target detecting algorithm, this thesis proposes a method that combining SVM and gradient feature to the moving object detecting. This method classifies moving target region and background region by SVM, and detects moving targets. The main work is to improve this method, improving the efficiency of moving object detecting. First of all, normalize the windows which calculating gradient features to 8 * 8. Next, normalize gradient value between 0 and 255, and express with 8 bytes binary number. Finally, in order to adapt to normalized gradient features, the SVM classifier is expressed in several binary base vectors approximately. By improving the SVM and gradient features, moving target detecting process could be like the process of binary operation, and improve computational speed.In terms of video summary technology which can make a long video condensed into a brief summary video. It is not only keeping original video information, but also can save storage space, convenient user locking interested information at the same time. Mainly, extract the key frames which have moving objects by the three frames difference method, and split the key frames out. Then fuse the moving object according to certainly density. Experiments show that this method is not only reduce the redundancy of video information, but also reduce the resource consumption of storage equipment, and to reduce the time that user brows video, allow the user to quickly understand the video content.