Multi-object tracking based on network flow model and ORB feature

Jieyu Chen, Zhenghao Xi, Junxin Lu, Jingjing Ji

Abstract: We present a multi-object tracking algorithm based on network flow model and ORB feature for solving occlusion problem. In our work, we extract the ORB features from the detection response and match them to achieve data association at the intra-tracklet stage. Then, tracklets are associated by network flow model to realize data association at the inter-tracklet stage. Each trajectory fragment is regarded as a node in the network flow model. Owing to the different occlusion situations, different network flow cost functions are proposed, by integrating the motion information obtained by the Edge Multi-channel Gradient Model, the appearance information of the tracklet and the time information. Experimental results demonstrate that compared with other state-of-art algorithms, our method improves tracking performance in complex environments.