Pothole Detection and Cost Estimation Using Deep Learning

Road damage due to potholes, etc., has a major effect on the safety of transportation systems as well as the management of the infrastructure. Traditionally, visually inspecting potholes can be both time-consuming and inefficient, often resulting in human error. This study describes an automated system designed to detect potholes and to develop a cost estimate for repairing them based on analysis of recorded videos using deep learning algorithms. The developed system uses an object detection algorithm called YOLOv8 to automate the detection of potholes from video input in real-time. The system also estimates the width and depth of detected potholes (and respective volume) with the aid of bounding box analysis and stereo vision. The estimate of the volume of the pothole also supports cost estimation of repairing the pothole by developing a list of materials required for repair. The developed system will be able to facilitate timely decision making by road maintenance authorities, enhance accuracy, reduce manual efforts, and improve the transparency and efficiency of resource allocation resulting in a better fit to be integrated into a smart city infrastructure. The experimental results indicate significantly improved detection accuracy, while at the same time providing satisfactory estimates of costs associated with repairing the damaged roadway (pothole detection) using this automated approach to both detecting potholes and estimating their cost, highlight the tremendous potential of the automated approach in modern-day road maintenance systems..