Robust multi-object tracking based on higher-order graph and min-cost flow network

Tracking-by-detection framework has been applied to most multi-object tracking algorithms, which consider the similarity measure between detection responses or tracks in a limited temporal window. This paper firstly construct the tracking model as a higher-order graph, exploiting higher-order information consistent with many objects in the temporal domain. And then the high order similarities between objects are applied in the flow network, thereby computing the optimal tracklets according to the minimum cost of the flow network. Extensive experiments on various challenging datasets of DETRAC, prove that the proposed algorithm performs have great performance in comparison with other methods.