Asset Condition Assessment Using AI and Computer Vision

Details

• Creators:
  Mashhadi, Ali ; Black, Clancy W. ; Wygonik, Erica
• Corporate Creators:
  Wall Consultant Group
• Corporate Contributors:
  Utah Department of Transportation ; United States. Department of Transportation. Federal Highway Administration
• Subject/TRT Terms:
  Artificial Intelligence Asset Management Barriers (Roads) Computer Vision Image Processing Machine Learning
• Publication/ Report Number:
  UT-25.12
• Resource Type:
  Tech Report
• Geographical Coverage:
  Utah ; United States
• Edition:
  November 2024 to June 2025
• Corporate Publisher:
  Utah Department of Transportation
• Abstract:
  UDOT faces significant challenges in manually inspecting roadside safety assets across its extensive roadway network, prompting the need for automated condition assessment solutions. This project leverages street-level image data to develop an AI-driven framework for assessing primary assets (guardrails, cable barriers, concrete barriers) and secondary assets (curbs, gutters, retaining walls, shoulder edge drop-offs). Using the YOLO11n model, object detection was evaluated across three settings—multiclass defect detection (mAP@50: 52%, mAP@50:95: 25%), binary classification (mAP@50: 41%, mAP@50:95: 15%), and multiclass with augmentation (mAP@50: 63%, mAP@50:95: 32%)—revealing poor performance despite improved results with augmentation, leading to the adoption of more advanced algorithms. Vision-language models (VLMs) were explored for contextual analysis, with testing conducted to assess both primary and secondary assets, focusing on defect classification, condition rating, and design compliance. Among VLMs, Gemma 3 offers the best accuracy (each image taking about 30 minutes on current GPUs), while Llama 3.2, with its advanced 11B parameter version, delivers the best performance in speed and evaluation. This framework creates significant opportunities to enhance UDOT’s maintenance prioritization, improving roadway safety and infrastructure efficiency.
• Format:
  PDF
• Funding:
  25-8370
• Collection(s):
  US Transportation Collection
• Main Document Checksum:
  urn:sha-512:90c3b37eb8e89b034016e53dec098e0bf9bbae1e3c98a91b774c3eb42146b59c62c998e97f4cb110c20b64c3e0101fecd3814253c022c9740adae025e0c204ac
• Download URL:
  https://rosap.ntl.bts.gov/view/dot/86003/dot_86003_DS1.pdf
• File Type:
  [PDF - 3.08 MB ]