RainyTrack: Enhancing Object Tracking in Adverse Weather Conditions with Siamese Networks
Jiacheng Liu · 2024
Object tracking is a critical task in the field of computer vision, playing a significant role in many practical applications. However, in complex rainy conditions, existing object tracking methods often perform poorly. This is due to rain introducing noise, blurriness, and occlusion in images, thereby reducing target visibility and tracking accuracy. Furthermore, the lack of large-scale training datasets specifically designed for rainy conditions limits the performance of existing methods in this environment. To address this issue, we propose a novel object tracker focused on achieving efficient and accurate object tracking in rainy conditions. We have enhanced traditional object tracking methods by introducing the "rain removal" concept and modifying feature fusion and attention mechanisms to improve the tracker’s performance in rainy conditions. Additionally, we synthesized rain on three commonly used object tracking datasets, LaSOT, OTB2015, and UAV123, incorporating varying degrees of rainfall and raindrop sizes to simulate real-world rainy conditions. Through extensive experiments on multiple publicly available datasets, we have validated the effectiveness of the proposed method. The experimental results demonstrate that our tracker excels not only on the original datasets but also maintains outstanding tracking performance on the rain-added datasets. This contribution provides new insights and methods for the advancement of the computer vision field.