Navigating Adverse Weather: An Exploration of YOLOv8's Object Detection in Fog

Object detection in adverse weather conditions, such as dense fog, poses a critical challenge for computer vision systems in applications like autonomous driving, surveillance, and robotics. Image quality degradation due to light scattering and absorption diminishes the performance of even the most advanced models trained on clear-weather datasets. This paper uses the Foggy Cityscapes dataset to evaluate YOLOv8, a cutting-edge object detection model, under extreme foggy conditions. We utilize YOLOv8's sophisticated architecture, featuring the CSPDarknet backbone and Path Aggregation Network (PANet), for efficient feature extraction and multi-scale detection. Our approach incorporates image enhancement techniques and data augmentation to mitigate the effects of fog on feature visibility. We conduct a comprehensive performance assessment using metrics like mean average precision (mAP) precision-recall curves to evaluate the model's robustness and reliability. Additionally, we benchmark YOLOv8 against previous YOLO versions, detailing improvements and trade-offs in detection speed and accuracy. We also propose a hybrid framework combining multi-task learning to improve image quality and object detection in low-visibility scenarios.