Online Object Tracking on Multiple Cameras with Completely Overlapping Views

This paper presents a tracking solution for multiple objects in a multi camera setup with overlapping fields of view. We utilize this not only to extend the overall operational area but also to improve tracking accuracy in the overlapping areas in case of occlusions and other practical issues. Our method lifts the usual bipartite approach to the multipartite domain using GMMCP which matches detections with their counterparts in all cameras and with their corresponding Kalman based track. This has two important properties, namely online and real-time capability, and it generates an optimal solution to the original problem without simplifications or approximations. It thus outperforms clustering and the bipartite methods as well as an offline method based on the k-shortest path algorithm.