Road pavement distress significantly affects traffic safety, vehicle durability, and environmental sustainability, making timely maintenance essential. Traditional inspection methods are labor-intensive, intrusive, and costly, while many data-driven approaches rely on expensive sensors or cloud computing, limiting scalability. This study presents a low-cost, non-intrusive monitoring system deployable on ordinary vehicles that performs automated pavement distress detection directly on-board using artificial intelligence. A dedicated image dataset was built by combining public resources with custom acquisitions, and pre-trained deep-learning models were adapted to this task. Three representative architectures were evaluated, including a lightweight real-time detector (YOLOv8n), a transformer-based detector (RT-DETR), and a two-stage detector (Faster R-CNN with a ResNet50 backbone). The recognition system was implemented on an embedded platform: NVIDIA Jetson Orin Nano. YOLOv8n achieved the best balance between accuracy and speed, processing 41 frames per second and achieving an average F1 score of 0.54 across all pavement distress classes. By linking edge artificial intelligence, pavement monitoring, safety, sustainability, and maintenance planning, the proposed framework supports an interdisciplinary approach to road infrastructure management. The results support the feasibility of affordable on-board artificial intelligence for road pavement assessment and its potential to prioritise targeted inspections and maintenance actions.
Petrongolo et al. (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: