Leveraging YOLOv8 and Geolocation API for Real-Time Pothole Detection

Computer vision applications have been transformed by deep learning, which makes it possible to recognize objects in a variety of domains with extreme accuracy and in realtime. Potholes are an extreme danger on roads, which increase the prospect of collisions, damaging cars, life threats and demanding more frequent maintenance. Prior research mostly employed YOLOv5, a real-time detection approach that strikes a balance between accuracy and speed. Nevertheless, there are still issues with increasing detection accuracy, decreasing false positives, improving real-time deployment. Training and assessment are conducted using datasets like RDD2022 and GRDDC2020. This research presents areal-time pothole detection system that makes use of YouOnlyLookOnce, Version8(YOLOv8), a cutting-edge deep learning model that is widely recognized for its rapidity and precision in object detection using video feeds from vehicle-mounted cameras, even in low-light conditions, fog, or rain. Experimental results demonstrate that our proposed YOLOv8 model achieves an accuracy of 92%, significantly outperforming YOLOv5, which achieved 78% accuracy. Our model also reduces false positives by 20% and improves detection speed by 35%. Additionally, accurate mapping of pothole sites is ensured by the Geolocation API, enabling automated road maintenance systems. Performance is evaluated using standard evaluation metrics, including Mean Average Precision (mAP), Intersection over Union (IoU), Precision, Recall, and F1-score.