Multiple Object Tracking by Joint Head, Body Detection and Re-Identification

Multi-object tracking (MOT) is an important problem in computer vision which has a wide range of applications. Formulating MOT as multi-task learning of object detection and re-Identification (re-ID) in a single network is appealing since it achieves real-time but effective inference on detection and tracking. However, in crowd scenes, the existing MOT methods usually fail to locate occluded objects, which also results in bad effects on the re-ID task. To solve people tracking in crowd scenes, we present a model called HBR (Head-Body-ReID Joint Tracking) to jointly formulates head detection, body detection and re-ID tasks into an uniform framework. Human heads are hardly affected by occlusions in crowd scenes, and they can provide informative clues for whole body detection. The experimental results on MOT17 and MOT20 show that our proposed model performs better than the state-of-the-arts.