Pothole Detection under Various Weather Conditions Using YOLOv9

Potholes pose significant dangers on roads, leading to substantial damage to vehicles, causing accidents, and increasing maintenance costs for authorities. Accurate and timely detection of potholes is crucial for effective road maintenance and repair. However, this task is challenging, especially under varying weather conditions. Rain, snow, and fog can obscure sensors and cameras, making it difficult to capture clear images of the road surface. Additionally, nighttime and shadows can further reduce the visibility of potholes. In this paper, we propose a pothole detection system that operates effectively under diverse weather conditions using YOLOv9, a real-time object detection algorithm. Our approach was trained and evaluated on a dataset of images generated under different weather scenarios. The resulting model achieved a precision of 0.807, a recall of 0.759, a mAP50 of 0.829, and a mAP₅₀-₉₅ of 0.551.