视频监控系统中的多摄像头跟踪优化设计

Multi-camera tracking optimization design in video surveillance system

针对分布式广域视频监控系统,提出了一种基于多摄像头融合的跟踪优化方法.该跟踪优化算法根据目标优先级和目标在各个摄像头中的遮挡状态及其分割图像大小进行数据加权融合,优先分配高优先级目标给具有最佳权值的摄像头进行跟踪,并动态平衡各个摄像头的跟踪负载,将跟踪负载过重的摄像头中的低优先级目标分配给其他摄像头进行跟踪.为了有效地建立重叠摄像头之间目标的对应关系,对于摄像头远离监控地平面和目标的场景,通过摄像头监视背景图像之间的SIFT特征匹配自动生成对应点,利用这些对应的关键点确定重叠摄像头之间的单应性变换矩阵参数,再根据目标质心坐标之间的单应性变换进行一致性匹配;对于摄像头近邻监控地平面和目标的场景,通过目标分割图像之间的SIFT特征进行一致性匹配.实验结果表明:该方法能有效地实现广域监控场景中多摄像头的协同跟踪,达到了较高的跟踪性能.

Aiming at the wide-area distributed video surveillance systems, a tracking optimization method based on multi-eamera fusion is proposed. This optimization algorithm prioritizes the fusion process based on the assigned priority, the occlusion state and image segmentation size of the moving target, and preferably allocates the optimal camera to track moving targets with high priority. The algorithm also dynamically balances the tracking load of each camera and allocates the high-load targets with low priority in the camera to other cameras. In order to effectively establish the correspondence between the targets of overlapping cameras, the target matching is implemented by homography transformation of target centroid coordinates in the scenes where the cameras are far away from the ground plane and targets, and by SIFT features matching of targets in other scenes. The coefficients of homography transformation arc computed by the corresponding keypoints created by SIFT features matching of surveillance background images of the cameras. Experimental results show that the method can effectively implement the cooperative tracking of multiple cameras in wide-area surveillance scenes, and achieves high tracking performance.