Video Object Segmentation Based On GA In Video Surveillance

Intelligent Video Surveillance System is a new intelligent digital video processing system. It has been widely applied in indoor and outdoor safety precautions of security agencies, urban road traffic control and so on. The video movement objects segmentation is the key technology in video surveillance and the first and foremost task of content-based video application. So, how to be effective in separating moving object from video images as well as the quality will directly affect the identification and tracking in intelligent video surveillance as well as video image compression. However, there are still many problems in respect to the current video objects segmentation algorithms, for example, the reliability and robustness of video objects segmentation need to be improved, algorithms were proposed in special application environment and the automatic video segmentation algorithm which can be used in any scene is still a classical difficulty which need to be solved. In order to better apply video objects segmentation to intelligent video surveillance system, more in-depth researches and finding a more effective algorithm about it are urgent and will be very important meaning not only in theory but also in reality.Based on the analysis results of previous studies, in this paper, we had a deep study and research to the video moving object segmentation techniques. To detect the movement regions, the method that edge detection combined two frames difference was employed in our algorithm. This method could avoid mistaking the background as the foreground caused by the occlusive and uncovered problems, simultaneously it was helpful to the suppression of stochastic noise influence, and could obtain the basic movement objects. In view of the moving object is easy to be affected by the noise, light, and so on, also has a high requirement in time and speed, on the basis of the existing algorithms we had discussed, an improved adaptive algorithm with the least error method was presented in the paper. The thesis implied an adaptive method to the cross probability and the mutate rate. In addition, we made use of the previous frames'information when we initialized the later populations, but not through the stochastic method, so the effect of noise was reduced and the real-time capability was greatly improved.Finally, we performed morphological filtering on the segmentation results, in order to eliminate the miscellaneous points and restore the moving foreground of the image sequences. At the end, it was proved that the method used in the paper not only acquired a good effect, but compared with some methods the real-time was advanced obviously.