Algorithm On Pedestrian Detecting And Tracking Across Multi-cameras Video Surveillance

With the advance of science, technology and living standard, People have paid more and more attention to public security. Due to the fact that the pedestrian tracking based on multiple cameras could make up the drawback of the pedestrian based on single camera, the algorithm on multi-cameras pedestrian tracking has been the hot pot for the further research. And it is attracting more and more attention. However, because of the interference of the strong illumination variations, very low resolution, pose and view-point changes and so on, the experts meet with many difficulties in the research on the pedestrian tracking based on multiple cameras. To solve these problems, this paper makes the improvement in the pedestrian detecting, pedestrian tracking and pedestrian handover. The detail is arranged as follows:First, we propose a pedestrian detection method based color invariance. we use the median function to model the background. After making use of the color invariance to achieve the detection of pedestrian, we use the appropriate filter function to detect pedestrians external contour processing and get a satisfactory target prospects, our algorism could handle the effect of the shadow.Second, we propose a pedestrian tracking algorism based on the color invariance and partial least square analysis, we use PLS to express the appearance feature of the pedestrian, and constantly update the mode. Our algorism could achieve the real-time effect and handle the occlusion.Third, we combine Retinex theory with the brightness to handle the influence of light, and reply to the original color.Forth, we propose the multi-camera handover algorism based on the topology of cameras and target re-identification. We set the distributed architecture of multiple cameras and use GMM to estimate the transition time. And we achieve the pedestrian re-identification by the iterative sparse representation. Our algorism could make sure the real-time effect of the handover tracking, and achieve the better matching ratio.