RaTrack: Moving Object Detection and Tracking with 4D Radar Point Cloud

Mobile autonomy relies on the precise perception of dynamic environments.Robustly tracking moving objects in 3D world thus plays a pivotal role forapplications like trajectory prediction, obstacle avoidance, and path planning.While most current methods utilize LiDARs or cameras for Multiple ObjectTracking (MOT), the capabilities of 4D imaging radars remain largelyunexplored. Recognizing the challenges posed by radar noise and point sparsityin 4D radar data, we introduce RaTrack, an innovative solution tailored forradar-based tracking. Bypassing the typical reliance on specific object typesand 3D bounding boxes, our method focuses on motion segmentation andclustering, enriched by a motion estimation module. Evaluated on theView-of-Delft dataset, RaTrack showcases superior tracking precision of movingobjects, largely surpassing the performance of the state of the art.