Artificial Intelligence for Pavement Condition Assessment From 2D/3D Surface Images

Details

• Creators:
  Wang, Feng ; Gong, Haitao ; Bai, Yongsheng ; Tesic, Jelena ; Luo, Xiaohua
• Corporate Creators:
  Texas State University--San Marcos
• Corporate Contributors:
  Texas Department of Transportation. Research and Technology Implementation Office ; United States. Department of Transportation. Federal Highway Administration
• Subject/TRT Terms:
  Accuracy Artificial Intelligence Assessments Automated Pavement Condition Data Collection Automation Continuously Reinforced Concrete Pavements Data Collection Data Management Data Quality Data Quality Threshold Datasets Dimensional Analysis Neural Networks Pavements Precision Quality Assurance Sampling

  More +
• Publication/ Report Number:
  FHWA/TX-26-0-7150-R1
• Resource Type:
  Tech Report
• Geographical Coverage:
  United States ; Texas
• Edition:
  09/2022-03/2026
• Corporate Publisher:
  Texas Department of Transportation. Research and Technology Implementation Office
• Abstract:
  The development of a national standard data format for two-dimensional/three-dimensional (2D/3D) pavement surface images and Artificial Intelligence (AI) Machine Learning (ML) in Computer Vision adopted by AASHTO and embraced by state DOTs has provided TxDOT the opportunity to develop new methods for automated pavement condition assessment. This research implements automated pavement condition evaluation using 2D/3D surface imagery and Artificial Intelligence. A comprehensive image library was established in the AASHTO standard data format, capturing diverse pavement conditions and surface types in Texas. The dataset includes 5,892 2D/3D image pairs and associated labels for Asphalt Concrete (ACP), 7,750 for Jointed Concrete (JCP), and 5,776 for Continuously Reinforced Concrete Pavement (CRCP). These correspond to 10,885, 16,943, and 13,779 individual distress instances, respectively, as defined by the TxDOT Pavement Management Information System (PMIS) for a total of 19,418 images. Neural network models such as YOLO (You Only Look Once) series were trained on the established library datasets for generalization and robustness of pavement distress measurements. Leveraging vision-based AI and ML models, the system automatically detects and measures surface distresses, achieving a mAP50 of over 0.80 for ACP, 0.70 for JCP, and 0.75 for CRCP on the validation datasets. These models were integrated into practical tools for calculating PMIS distress scores and delivered via a dedicated software application for TxDOT. A pilot study was implemented using both the library and vendor-collected field data in 2024 to validate the efficacy of the proposed methods. The research concludes with actionable recommendations and a strategic roadmap designed to transition these findings into full-scale operational implementation within TxDOT’s pavement management system.
• Format:
  PDF
• Funding:
  0-7150
• Collection(s):
  US Transportation Collection
• Main Document Checksum:
  urn:sha-512:b43667b6217cce22856fe63357b18c51b32ef43420fb08e8e94adac80820e1b379e533236246dd30cfb99f1f21ac4cbd2c20980d38719921655938b882da68f6
• Download URL:
  https://rosap.ntl.bts.gov/view/dot/90511/dot_90511_DS1.pdf
• File Type:
  [PDF - 16.19 MB ]