AI for Automated Asphalt Pavement Distress Assessment from Vehicle-Mounted Imaging

This course examines artificial intelligence applications for automated asphalt pavement distress assessment. Topics include the federal regulatory context and distress taxonomy, data acquisition modalities including 2D and 3D imaging systems, computer vision architectures for classification, object detection, and semantic segmentation, and the factors that affect model performance in agency practice. Students will gain the technical knowledge to specify, accept, and responsibly rely on AI-derived pavement condition data in engineering deliverables.

Learning Objectives

1. Identify the federal performance measure requirements under 23 CFR Part 490 and the national consensus standards (AASHTO R 85, AASHTO R 86, ASTM D6433) that govern automated pavement condition data and their key technical performance criteria.

2. Distinguish the three computer vision task categories used in pavement distress recognition (classification, object detection, and semantic segmentation) and explain why pixel-level segmentation is required for engineering-grade distress quantification.

3. Describe the principal factors that affect AI model performance in pavement applications, including training data quality, domain shift, annotation quality, and the limitations of vendor-reported metrics on private datasets.

4. Apply the specification and acceptance requirements for automated pavement condition surveys, including performance criteria, ground truth establishment procedures, and documentation of data limitations in engineering deliverables.