Multi-Object Tracking Based on Deep Path Aggregation Network
Xi Chen, Yifeng Zhang · 2022 7th International Conference on Image, Vision and Computing (ICIVC) · 2022
Multi-object tracking is an important task in the field of computer vision. With the rapid development of object detection and re-identification, object tracking performance has been improved accordingly. It has been proved that FairMOT jointly performs object detection and re-identification tasks in a single network and obtains a high level of detection and tracking accuracy. Based on the DLA model, this paper proposes a novel tracker equipped with a Deep Path Aggregation Network (DPANet) to effectively improve multi-object tracking accuracy. The network enhances the entire feature hierarchy with accurate localization signals in lower layers by bottom-up path augmentation. In the final stage, we further concatenate multiple feature maps in parallel to maintain rich high-resolution representations. Experimental results show that the tracker outperforms the most previous state-of-the-art on several public datasets.