An Efficient Multiple Object Vision Tracking System using Bipartite Graph Matching

For application domains like 11 vs. 11 robot soccer league, crowd surveillance and air traffic control, vision systems need to be able to identify and maintain information in real time about multiple objects as they move through an environment using video images. In this paper, we reduce the multi-object tracking problem to a bipartite graph matching and present efficient techniques that compute the optimal matching in real time. We demonstrate the robustness of our system on a task of tracking indistinguishable objects. One of the advantages of our tracking system is that it requires a much lower frame rate than standard tracking systems to reliably keep track of multiple objects.