FATrack: Frequency Alignment for Robust Visual Object Tracking in Hazy Conditions

Visual object tracking methods for hazy conditions have advanced rapidly in recent years, effectively alleviating the pronounced performance degradation of conventional trackers caused by reduced contrast and loss of fine details in hazy conditions. However, most existing hazy condition tracking methods focus on feature extraction in the spatial domain, with limited exploitation of frequency domain information. Even when frequency domain features are introduced, the role of different frequency bands is rarely modeled explicitly, making it difficult to obtain stable and highly discriminative target representations when details are degraded in hazy scenes. To address this issue, we propose FATrack (Frequency Alignment Tracker), which incorporates frequency domain information while preserving spatial domain features. It constructs a multi-branch feature mapping that jointly models spatial and frequency representations and explicitly captures the complementary contributions of different frequency bands, thereby enabling robust object tracking in hazy conditions. Specifically, we design a Frequency Representation Alignment (FRA) module that enhances robustness under haze by enforcing consistency of frequency representations. In addition, we propose an Enhanced Consistency Distillation (ECD) strategy, which imposes a consistency distillation constraint using supervision signals from augmented views to mitigate response shifts induced by degradation. Extensive experiments demonstrate that FATrack achieves superior performance to mainstream state-of-the-art trackers on hazy visual object tracking benchmarks.