This study presents an AI-enabled framework for automated pavement condition assessment in urban environments by integrating YOLOv8-based distress detection, computational Pavement Condition Index (PCI) estimation, and comparative validation against manual PCI inspections and Pavement Surface Evaluation and Rating (PASER) scores. A YOLOv8 object-detection model, implemented in Python and trained on the publicly available N-RDD2024 dataset, was developed to identify longitudinal cracks, transverse cracks, alligator cracking, and potholes. The model achieved an accuracy of 84.6%, a precision of 89.6%, and a recall of 86.3%, demonstrating robust detection performance under heterogeneous environmental conditions. Dash-cam imagery collected along 6.3 km of urban flexible pavements was processed through an automated workflow that detects pavement distresses, estimates their severity and extent, and computes PCI values according to ASTM D6433-20 procedures. Automated PCI values were compared with manual PCI inspections and PASER ratings generated by the Blyncsy platform across 23 pavement sections. Statistical validation between automated and manual PCI assessments returned an R-squared of 0.925, a Pearson correlation coefficient of 0.962, a Spearman correlation coefficient of 0.955, a Mean Absolute Error of 5.0 PCI points, and a Root Mean Square Error of 6.1 PCI points. Compared with the proposed framework, PASER ratings exhibited lower agreement with manual PCI assessments and generally overestimated the pavement condition. The results demonstrate the potential of low-cost AI-based systems for large-scale pavement monitoring. Nevertheless, performance degradation was observed under challenging environmental conditions and in heavily deteriorated sections, highlighting the need for improved distress quantification, dataset balancing, and multimodal sensing integration.
Serrone et al. (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: