Abstract:

Multi-target tracking is an important research direction in the field of computer vision. Multi-target tracking plays an important role in the fields of intelligent video surveillance, human-computer interaction, robot navigation, and public safety. At present, the target tracking algorithm still faces many challenges, such as the effects of occlusion, complex background, motion blur, etc., which are difficult to completely avoid. This paper proposes an algorithm that combines multiple types of information, which effectively improves the performance of the tracker. The model focuses on the problem of object detection and data association between frames, which depends on the similarity of the target motion and the apparent between different frames. When the target is lost and there is occlusion, the fusion of multi-source information reduces the related uncertainty. At the same time, the algorithm achieves real-time tracking performance in real-world environments. Experimental evaluation shows that the proposed tracker has good performance on the public data set, greatly reducing the target loss rate and ID switch.

Key words: computer vision, multi-target tracking, object detection, target occlusion, real-time tracking, data association