Real-Time Pothole Detection Using Deep Learning and YOLO Variants

Road infrastructure deterioration, particularly in the form of potholes, poses a significant threat to global road safety, causing severe traffic accidents and substantial vehicle damage. Traditional methods for identifying road anomalies rely heavily on manual inspections and citizen reporting, which are inherently inefficient, labor-intensive, and prone to human error. To overcome these limitations, this research proposes an automated real-time pothole detection system leveraging deep learning architectures. Specifically, this study evaluates and compares the performance of YOLO variants, including YOLOv5, YOLOv7, YOLOv8, and YOLOv11, alongside the proposed YOLOv26n model. The models were trained and fine-tuned on an annotated, augmented dataset of diverse road-surface images to ensure robustness across varying environmental and lighting conditions. Experimental results demonstrate the efficacy of the proposed approach, with YOLOv26n achieving a mean Average Precision (mAP) of 89.1%, an overall detection accuracy of 87.4%, and a real-time inference speed of 42 FPS. This intelligent detection framework empowers municipalities to proactively prioritize road maintenance and enhance overall vehicular safety.