Enhancing Road Damage Detection Performance using the YOLOv9 Model

Abstract

Roads are essential infrastructure that support community mobility, and their condition has a significant impact on road user safety. However, manual road damage detection remains inefficient, time-consuming, costly, and prone to human error. To address this issue, this study proposed the YOLOv9 model for automated road defect detection, with an additional exploration of parameter combinations to optimize its performance. The proposed solution leverages the YOLOv9 model, which offers enhanced detection speed and accuracy compared to previous YOLO versions, due to its improved backbone and dynamic label assignment techniques. The method utilizes pre-trained weights and conducts parameter tuning to adapt the model to the task of identifying common road defects, including potholes, longitudinal, lateral, and alligator cracks. A publicly available road condition image dataset was utilized for training and evaluation purposes. Experimental results demonstrated that the optimized YOLOv9 model achieved a precision of 76.2%, a recall of 49.2%, and a mean Average Precision (mAP) of 62.8%, indicating a promising capability in accurately detecting multiple types of road damage. This study highlights the potential of YOLOv9 as an effective tool for road monitoring systems, contributing to proactive maintenance strategies and more efficient infrastructure management.