YOLOv11-s-Based Pothole Detection Model with Integrated Traffic Data for Maintenance Decision Support

Abstract

A pothole is a structural failure in a road that is a major cause of disturbances to traffic flow, particularly in developing countries where infrastructure maintenance is often delayed or inefficient. Traditional manual inspection methods for identifying potholes and planning repairs can be time-consuming. This study presents an automated pothole detection model using the latest YOLOv11-s object detection algorithm. Unlike previous research that focused on driver alerts and pothole classification based on the size of the pothole, this model integrates pothole detection with a traffic data API to prioritise road repair works based on traffic volumes. The dataset used for training, validation and testing was obtained from the RoboFlow platform, while images from Sri Lankan roads were used for deployment validation. The model achieved a precision of 0.851, a recall of 0.676, and mean average precision (mAP) scores of 0.802 (IoU@0.5) and 0.510 (IoU@0.5:0.95), demonstrating strong detection capabilities. The developed model can assist maintenance authorities in prioritising road repairs based on traffic volume. It can be further enhanced by integrating pothole characteristics, such as type and size, with traffic data to improve decision-making on the priority of repair works.