Abstract
Abstract
Autonomous vehicles encounter serious difficulties in adverse weather conditions such as fog, snow, rain, and sandstorms, where image quality is reduced and object detection becomes less reliable. Many existing approaches perform poorly in such environments because they are trained mainly on clear-weather datasets and rely on generic preprocessing methods. This study presents a framework that combines weather-based preprocessing, image enhancement, and optimized object detection. Weather conditions are first identified using ResNet-18. Based on the detected condition, the right preprocessing methods, like dehazing and denoising, are used to make the image clearer. ESRGAN is then used to improve the processed images by bringing back important details that are needed for detection. Bayesian optimization is used to make the YOLOv8 model better at finding objects in real time. The results show improved performance across different weather conditions. The model achieves precision values of 81.08% in snow, 85.64% in rain, 85.44% in fog, and 65.76% in sand. The corresponding mAP values are 75.95% for snow, 70.80% for rain, 75.35% for fog, and 49.50% for sand. These results show that the system can still find things even when conditions change. The framework makes it easier to find objects in difficult places and makes it safer for self-driving cars to work.