Learning Spatio-Temporal Information for Multi-Object Tracking

The robust multi-object tracking problem is a challenging issue in the field of computer vision. In this paper, we propose a multi-object tracking algorithm with temporal-spatial information and trajectory of confidence. The whole process is divided into local and global association. Trajectories with high confidence are associated with the detection result of the current frame during local association, whereas trajectories with low confidence are associated with the detection results of the current frame are not matched during global association. We determine the association results using a combined model. By utilizing the information of spatial-temporal correlation, the model is more robust and can deal with missed detection. In addition, we measure the reliability of the spatial information by the confidence map smoothing constraint and the peak sidelobe ratio criterion. We conduct experiments using a challenging public data set, and the results show that our proposed algorithm is superior to many other popular algorithms when dealing with problems, such as missed detection and poor tracker robustness.