Enhanced Transformer-Based UAV-to-UAV Tracking Using Distractor Elimination and Trajectory Prediction

With the advancement of UAV technology, utilizing UAVs for surveillance, capture, or destruction of other UAVs has become a significant focus in the counter-UAV domain. A core challenge within this domain is the visual target tracking of a UAV by the other UAV. However, due to the highly dynamic relative motion between the source UAV and the target UAV, along with the small scale of the UAV in images and background distractors, existing tracking algorithms, including the most popular Transformer-based trackers, often struggle to continuously and robustly track the target. To address these issues, this paper proposes an enhanced Transformer-based UAV-to-UAV tracking method using distractor elimination and trajectory prediction. Specifically, a target-surrounding matching mechanism is introduced to help the tracker eliminate distractors. Additionally, to tackle the high dynamic tracking problem of the camera and target, a Transformer-based trajectory prediction module is proposed to accurately predict the target’s location even when the tracker loses the target. Experimental results demonstrate that the proposed method achieves superior performance in UAVto-UAV target tracking scenarios compared to state-of-the-art algorithms