Research On Multi-target Detection And Tracking In Video Surveillance

With the research of intelligent monitoring system becoming embedded and universal, the detection and tracking of moving targets is become a research hotspot in the computer vision field. The research of moving object detection and tracking still faces a large number of challenging issues as the the change of light, complex background, diverse shapes, mutual occlusion of many objects and so on. This paper mainly studies the detection and tracking of multiple moving targets in static scene technology. The main work is as follows:About the object detection, we find the target area by VIBE algorithm which is optimized by Otsu guidelines; we remove the ghost in VIBE algorithm. First of all, make the VIBE to get the judge threshold adaptively through OTSU rule, then we get the optimal result in the prospects of the judge and the object segmentation. Secondly, we use the HOG SVM classifier based on the results to detect after treatment, thirdly, we detect and compare the contour similarity of foreground and background by VIBE to determine whether the foreground is ghost or stay object, remove the foreground if it is ghost or stay object in different way. The methods mentioned above can detect and segment the moving object accurately.About the object tracking, The influence of occlusion on tracking and the bad real-time are the two disadvantages of Multi target tracking algorithm based on trajectory optimization. This paper presents the optimization of tracking and improve the real-time of the method. In this paper, the increasing coefficient of persistent option of energy function to increase the ability of track fusion, In order to prevent the emergence of false track fusion, also joined the compare spacer frame in the energy function, thereby enhancing the ability of long time tracking occluded target. Then, segment the video sequence and tracking using trajectory optimization method, connected in the interface between section and section. So we don’t have to wait until all the target are detected completely, this way enhanced real-time visually. According to the experiments done by this paper, the above methods are feasible and effective.