City Road Condition Evaluation Through Artificial Vision Information When Driving: A Machine Learning Approach

The condition of urban roads is a major concern for many city councils, as it directly impacts the safety of both vehicles and pedestrians. Maintaining roads in optimal condition necessitates regular and thorough inspections, which can be both costly and resource-intensive. This study introduces a novel approach to monitoring road conditions by employing a fusion strategy that combines camera and instance segmentation techniques, enabling precise pothole identification and mapping within the urban landscape. In our research, we have evaluated the performance of four iterations of the YOLO (You Only Look Once) algorithm (YOLOv5, YOLOv6, YOLOv7, and YOLOv8) for the detection and characterisation of potholes. The systems were trained using publicly available datasets. Our findings reveal the strengths and weaknesses of each YOLO version in pothole detection tasks, with a surprising outcome: YOLOv5 demonstrates superior performance over its more recent successors. This work not only provides a feasible solution for continuous and automated road condition monitoring but also offers a comparative analysis of cutting-edge detection technologies, contributing valuable insights for future urban infrastructure maintenance strategies.