CSRTrack: a cascaded searcher with dynamic target recognition for intelligent visual sensing and tracking

With the rapid development of UAV platforms and high-resolution vision sensors, achieving reliable object tracking across diverse sensing environments has become a crucial requirement for intelligent visual sensing systems. However, object disappearance—caused by occlusion, viewpoint changes—remains a major obstacle that significantly compromises tracking stability and robustness. Most existing methods lack effective strategies for handling disappearance and for resuming tracking when the object reappears. To address these issues, we propose CSRTrack—a cascaded searcher with dynamic target recognition—that enhances a tracker’s ability to handle target disappearance and reappearance. First, we design a presence checker to determine whether the target has disappeared, thereby triggering the subsequent cascaded searcher. In addition, because disappearance typically arises from leaving the field of view or from complete occlusion, we adopt a cascaded search strategy to improve tracking failure recovery under these conditions. Moreover, we introduce a dynamic target-recognition module that improves adaptability to appearance variations. Finally, we evaluate our approach on four benchmark datasets; in particular, on UAV20L, our method improves the success rate by 19% and precision by 20% over the baseline, demonstrating competitive tracking performance.