Automated Road Damage Detection and Interactive Mapping Using YOLOv11, YOLOv12, and DeepSORT

Abstract

Road damage significantly impacts infrastructure efficiency, road safety, and maintenance budgets. This paper introduces an automated road damage detection and visualization system employing advanced deep learning models, specifically YOLOv11 and YOLOv12 [1], [2], with ongoing testing of DeepSORT [3] for improved detection tracking across video frames. Utilizing a Raspberry Pi [6] equipped with a camera and GPS module, synchronized video and GPS data are captured. Data is uploaded to a Node.js [8] and Next.js [9] web platform for processing, resulting in an interactive, color-coded map allowing detailed damage analysis and route navigation based on road damage data. Our model achieves a mean Average Precision (mAP) of 54%, indicating significant practical applicability.