Non-Maximum Suppression Performs Later in Multi-Object Tracking

Multi-object tracking aims to assign a uniform ID for the same target in continuous frames, which is widely used autonomous driving, security monitoring, etc. In previous work, low-scoring box, inevitably contained occluded target, was filtered by Non-Maximum Suppression (NMS) detection stage with confidence threshold. order track effectively, this paper, we propose method of NMS performing later. The works rather than stage. More candidate boxes that contain are reserved trajectory matching. addition, unrelated discarded according Intersection over Union (IOU) between predicted and detected box. Furthermore, an unsupervised pre-trained person re-identification (ReID) model applied improve domain adaptability. bicubic interpolation increase resolution boxes. Extensive experiments on MOT17 MOT20 datasets have proven effectiveness targets proposed method, achieves MOTA 78.3%.