Research And Implementation Of Robust Motion Detection Under Specific Environment In DM642 Based Intelligent Video Surveillance

With the development of artificial intelligence and computer science, Machine Vision Technology has made considerable progress, and attracted more and more attention. Intelligent video surveillance is an advanced application based on Machine Vision Technology, which is able to detect and identify different objects, discover abnormal situations then to trigger alarm events in the fast and the best ways. Nowadays, the intelligent video surveillance has served as a powerful auxiliary tool to fight against terrorism. In addition, the intelligent video surveillance can also be used in traffic management, customer behavior analysis, customer service and other non-security-related scenarios.However, the different complex environments have always been giving rising challenge to the intelligent video surveillance technology. The focus of the paper is on the research of motion detection and robust background modeling algorithms under specific environments together with the realization of the DM642 based intelligent video surveillance system.In this paper, two different background modeling algorithms are advanced to deal with two specific environments. To effectively distinguish between ghosts and left objects under surveillance videos including moving objects from the beginning, a spatial-temporal correlation based background modeling algorithm (TSCEP) is proposed. TSCEP can fast eliminate the negative effect of ghosts and still identify left objects as foreground. The other algorithm (MSCG) combines the advantages of Background Difference method and GMM method and robustly achieves illumination invariant motion detection under most complex environments.TI's DSP DM642 is mainly designed for digital video's system solutions. After the transplantation and optimization of the algorithms on DM642 platform, an intelligent video surveillance system is designed to meet different surveillance applications. Experimental results over a variety of video sequences have shown that the implemented algorithms could achieve good performance in motion detection and surveillance applications. And real time processing is achieved for videos, which can satisfy the demand of surveillance applications.