Dist-Tracker: A Small Object-aware Detector and Tracker for UAV Tracking

Wenzhen Wang, Jing Fu, Jiayi Song, Kaiyu Li, Hui Qiao, Jiang Liu, Hao Sun, Xiangyong Cao; Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR) Workshops, 2025, pp. 6601-6609

The widespread adoption of civil unmanned aerial vehicles (UAVs) has accelerated the development of anti-UAV technologies. Despite thermal infrared video enables all-weather surveillance, existing methods for multi-UAV tracking struggle with low thermal contrast, object scale variation, and erratic motion patterns. In this paper, we propose Dist-Tracker, a two-stage framework integrating a Scale-Shape-Quality (SSQ) detector based on YOLOv12 and Fusion of L2-IoU Tracker (FLIT) to address these challenges. For detection, SSQ introduces scale-aware Wasserstein distance with covariance alignment, dynamic shape-aware penalties, and adaptive gradient modulation to resolve geometric instability in small infrared targets. For tracking, FLIT synergizes IoU and L2 metrics with camera motion compensation, mitigating spatial jitter and occlusion-induced ambiguities through hybrid cost metric optimization. Comprehensive evaluations on the validation set from the Anti-UAV dataset demonstrate that our proposed framework achieves remarkable performance, with a detection AP50 of 93.9% and a tracking MOTA of 77.5% in cluttered infrared environments, significantly advancing UAV swarm detecting and tracking capabilities through geometric-stable perception and motion-resilient association. Our method won first place in the 4-th Anti-UAV challenge Track3 (tracking MOTA:81.32% on the official test set).