Constraints-based graph embedding optimal surveillance-video mosaicing

One single surveillance camera usually possesses a local view point, making it incapable of monitoring the whole scene. Even utilizing a pan, tilt, zoom (PTZ) platform, the captured frames are inconvenient for observation due to its discretion in presenting the global area. In this paper, we focus on the video mosaicing problem in surveillance environment, to provide a convenient and accurate panorama of the global site being monitored. Based on the specific requirements in surveillance environment, three constraints are proposed for video mosaicing, which include temporal constraint, spatial constraint and similarity constraint. Furthermore, we cast the problem of video mosaicing into the framework of graph embedding, and solve the optimum mosaic by finding the optimal path in the graph with Dijkstra method. Experiments from both outdoor and indoor applications demonstrate the advantage of our method over previous ones, and show that our method is very adaptable in surveillance applications.