Real-Time Detection of Road Defects Using YOLO Architectures: A Comparative Study

Abstract

Road surface degradation presents significant challenges for infrastructure maintenance and traffic safety. This study presents a comparative evaluation of two deep learning models, YOLOv5 and YOLOv8, for the automatic detection of pavement defects including longitudinal cracks, transverse cracks, alligator cracks, and potholes. A custom dataset of high-resolution images was manually annotated and used to train both models under identical conditions. The evaluation includes precision-confidence curves, recall-confidence curves, and precision-recall curves, as well as a visual comparison of detection results. YOLOv5 demonstrated superior accuracy, consistency, and reliability, especially in detecting subtle defects such as longitudinal cracks. YOLOv8, despite its recent architecture, exhibited greater variability in prediction confidence and bounding box precision. These results indicate that YOLOv5 is a strong candidate for real-time road surface monitoring. Further investigation will be aimed at improving detection across a range of environmental conditions and continuing evaluation on larger and more varied data sets.