A Multi-View Pedestrian Tracking Framework Based on Graph Matching

In the applications of video monitoring over large public or private spaces, multiple cameras are required to cover the entire space and resolve the problems of occlusion, object intersection and so on. In this work, a novel multi-view pedestrian tracking framework is proposed to simultaneously detect and associate human objects across views using graph matching techniques to fully exploit the object features and the spatial/temporal relationships among the objects. Experimental results are provided to demonstrate the accuracy of our proposed framework.