Robust Object Detection Based on YOLOv8 by Snow Removal Augmentation Method

In adverse weather conditions, particularly in snow, autonomous driving systems face significant challenges in accurately detecting objects like vehicles and pedestrians. This paper presents an enhanced snow detection model based on YOLOv8, integrating a snow removal module to improve detection performance in snowy conditions. The model addresses the challenges of snow-induced noise through data augmentation and snow removal techniques, optimizing YOLOv8 for various intensities of snowfall. Contributions include the development of a headlight detection approach that utilizes HSV color space adjustments to separate headlight regions from snow noise, preserving critical features during snow removal. Experimental results demonstrate that YOLOv8 with the snow removal module performs significantly outperforming baseline models like YOLOv5 and YOLOv11 in identifying vehicles and road elements in snowy environments. These improvements enable the system to maintain higher accuracy and robustness in adverse conditions.