A Novel LiDAR-Camera Fused Player Tracking System in Soccer Scenarios

Localization and tracking in multi-player sports present significant challenges. Existing GPS and UWB-based methods have higher accuracy but require players to wear inconvenient electronic devices. On the other hand, camera-based solutions are device-free, but have unsatisfied accuracy and necessitate significant manual intervention. Thus, we introduced a LiDAR-Camera fused system tailored for soccer scenarios, that can combine highest position accuracy and convenient application. With this continuous data collection system, a new benchmark dataset is constructed, and a simple and effective tracking-by-detection method is designed. The method comprises two main module. The Cascade PointNet module effectively identifies players from the point cloud, while the subsequent Tracklet-Zoom module efficiently performs tracklet association. Our method shows significant improvements in multiple evaluation metrics compared to other methods. Specifically, we achieved a score of 95.21% on the GT-consistency metric, representing a 13.29% increase. This indicates that our method has made substantial progress towards achieving automated tracking.