Learning disturbance-aware correlation filter with adaptive Kaiser window for visual object tracking

Abstract

Discriminative Correlation Filters (DCF) have been recognized as a classic and effective method in the field of object tracking. In order to mitigate boundary effects, prior DCF-based tracking methods have commonly employed a fixed Hanning window, limiting the adaptability to fluctuations of the response map. Therefore, we propose a disturbance-aware correlation filter with adaptive Kaiser window (DCFAK) for visual object tracking. The adaptive Kaiser window dynamically adjusts its values according to the kurtosis of the response map, effectively suppressing boundary effects. Additionally, to further improve robustness, our DCFAK introduces a disturbance peaks suppression method, which can better distinguish the target object from the objects with similar appearance in the background by attenuating the sub-peaks within the response map. We comprehensively evaluate the performance of our DCFAK on seven datasets, including OTB-2013, OTB, 2015, TC-128, DroneTB, 70, UAV123, UAVDT, and LaSOT. The results demonstrate the superior performance of our method across these datasets.