Discriminative correlation filter (DCF)-based trackers are widely used in the field of UAV target tracking due to their high computational efficiency. However, existing DCF trackers suffer from a limitation where tracking drift occurs because they incorrectly update the filter with background information during occlusion. In this paper, we propose a robust tracking method applying confidence-based re-detection algorithm to address this problem. The proposed method evaluates tracking confidence using a multi-metric mechanism that integrates APCE (Average Peak-to-Correlation), PSR (Peak-to-Sidelobe Ratio), and Fmax(Maximum Response) to precisely detect occlusion. Following occlusion events, the target is re-detected using a backup model stored during high-confidence tracking. Furthermore, the searching window is positioned by considering both the target’s estimated location, which reflects its inertia during normal tracking via an exponential moving average (EMA), and occlusion location. To address the position uncertainty without resolution loss, an adaptive multi-search window strategy is employed instead of simple scale expansion. Also a post-redetection validation mechanism is applied to enhance tracking stability. Experiments using the UAV123 dataset and VisDrone2019 confirm that the proposed method demonstrates superior re-detection performance and robustness compared to existing methods in occlusion scenarios.
Lee et al. (Mon,) studied this question.