Robust Multi-Object Tracking with Local Appearance and Stable Motion Models

Multi-object tracking (MOT) has been steadily studied for video understanding in computer vision. However, existing MOT frameworks usually employ straightforward appearance or motion models and may struggle dynamic environments with similar complex motion. In this paper, we present a robust framework local stable to overcome these two hindrances. The incorporates object part detectors, feature extractor, keypoint data association method. For the association, utilize five types of similarity metrics cascaded matching strategy. model is suggested be used additionally global features full bounding boxes obtain discriminative even objects appearance. At same time, considers core body as central point subdivides using novel 12-tuple Kalman state vector analyze As result, our new tracker achieves state-of-the-art performance on DanceTrack test set, surpassing all other listed systems terms both detection quality metrics, including HOTA, DetA, AssA, MOTA. source code available at https://github.com/Jubi-Hwang/Robust-MOT-with-Local-Appearance-and-Stable-Motion-Models.