Lane Tracking for Autonomous Road Pavement Inspection with Unmanned Aerial Vehicles

Pavement distress assessment is a critical aspect of road infrastructure maintenance and safety. Traditional methods relying on manual surveys have proven time-consuming, labor-intensive, costly, and prone to human errors. Emerging developments in computer vision and deep learning techniques have opened new avenues for automated pavement distress detection using drones. Autonomous road inspection scheme using UAVs requires intelligent detection and tracking of road lanes. In effect, the navigation of the UAV will not require direct control from the human operator. This paper highlights the efficiency of road lane tracking using the instance segmentation approach of the You Only Look Once (YOLOv8) architecture. It also introduces a dataset of UAV images from complex environments within the highway roads in Northern Mindanao of the Philippines. The trained YOLO model garnered 99.98% precision, 80.12% recall, and 87.21% mean average precision. Its network performance showed a reliable self-correcting navigation mechanism for the autonomous UAV. The proposed approach could aid the UAV in autonomous maneuvering during inspection. Thus, future work includes experimentation with the feedback control system of the UAV where the road lane tracking is one component.