Multi-object tracking by virtual nodes added min-cost network flow
Peixin Liu, Xiaofeng Li, Haoyang Feng, Zhizhong Fu · 2017
In this paper, we propose a Virtual Missing-detection and Occlusion Model (VMOM) for the cost flow network approaches in multiple object tracking (MOT) systems. The VMOM generates virtual nodes which represent occlusions and missing detections. Regular nodes which indicate tracklets together with virtual nodes are used to establish a cost flow network. A Counter Embedded Iterative Shortest Paths (CEISP) algorithm is also proposed to solve the new network flow optimization efficiently. Finally, the tracker with our scheme is tested on the open Multiple Object Tracking Benchmark. Experimental results show the advantages of our approach in main performance indicators comparing with many excellent trackers.