StrongSiamTracker: A Siamese Tracker with Dynamic Global Detection for Robust Anti-UAV Tracking

Xiaolong Cui, Liu Wan, Lingqi Kong, Jimin Li, Chaojie Zhang, Ruohan Zhao, Panlong Wu, Shan He; Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR) Workshops, 2025, pp. 6619-6629

In the field of anti-UAV tracking, researchers often face challenges such as scale variations, background interference, and rapid movement, all of which significantly impair tracking performance. To address these issues, we propose a Strong Siamese Drone Tracker that dynamically integrates global detection and local tracking branches. In the initial tracking phase, we have designed a detector specifically tailored for small-sized targets to swiftly locate the drone's position. This detector incorporates a dynamic detection head to enhance the feature representation of small targets and employs a loss function based on Normalized Wasserstein Distance to more accurately assess the similarity between small targets. To mitigate short-term target loss caused by distractors in the tracking branch, we have integrated a trajectory-guided Spatial Constraint Proposal Suppression strategy into the baseline framework. This strategy selectively retains candidate regions that are spatially consistent with the target's historical motion trajectory. Additionally, we propose an adaptive Kalman filter that predicts target states during tracking failures. Finally, we establish dynamic switching rules between the two branches. Extensive experiments conducted on the 4th Anti-UAV Challenge and the Anti-UAV410 dataset demonstrate the superiority of our method. Notably, we achieved fourth and third places in Track 1 and Track 2 of the 4th Anti-UAV Challenge, respectively.