Enhance Bridge Image Attribution Through Automated Post Image Processing

Details

• Creators:
  Attanayake, Upul ; Battu, Naveen ; Chun, Pang-jo ; Weerasinghe, Rajith
• Corporate Creators:
  Western Michigan University. College of Engineering and Applied Sciences
• Corporate Contributors:
  Michigan. Department of Transportation ; United States. Department of Transportation. Federal Highway Administration
• Subject/TRT Terms:
  Artificial Intelligence Asset Management Bridges Computer Vision Data Segmentation Defect Detection Defects Detection And Identification Systems Deterioration Feature Identification Image Analysis Image Processing Inspection Literature Reviews Object Detection

  More +
• Publication/ Report Number:
  SPR-1771
• Resource Type:
  Tech Report
• Geographical Coverage:
  United States ; Michigan
• Edition:
  Final Report: 09/01/2024 – 02/28/2026
• Corporate Publisher:
  Michigan Department of Transportation. Research Administration
• Abstract:
  Highway agencies and consultants capture thousands of images during biennial, scoping, and request for action (RFA) inspections. Most of these files are stored with limited descriptions or inconsistent naming formats. As a result, a rich data source is underutilized. Understanding the value of this rich data source for asset management, MDOT initiated this project to explore how computer vision and artificial intelligence (AI) can automatically organize, label, and analyze bridge inspection images, turning unstructured images into structured data that can directly support the asset management program. The comprehensive review of state-of-the-art literature and practice showed that the commercially available tools are limited to roadside asset management. None of those tools is capable of performing semantic segmentation to detect bridge components. The foundation models, such as OpenAI’s GPT-5.1 and Google’s Gemini 3 Pro, are capable of delivering descriptive answers to given prompts. However, these models are not error-proof; hallucinations, non-determinism, and the reproducibility of results remain major issues when using them for highly specific and complex tasks such as bridge image analysis. The capabilities of the Integrated Bridge Analysis System, a comprehensive model for performing semantic segmentation and damage detection of bridge components, were demonstrated. Recommendations from this project include integrating the demonstrated model into the structural inspection program as a standalone application and using it to batch-process a large volume of images in the inspection database, converting them into a rich data source.
• Format:
  PDF
• Funding:
  2022-0434z7
• Collection(s):
  US Transportation Collection
• Main Document Checksum:
  urn:sha-512:a95dc5953aa8c48b855a9c97c52fed2e4a2688bbd99fbde90a5c393d67e00d0e90e7b21f7c6838ec809619986ec4ee62bcdf2fcadd22a9a6221dcd4857fa3281
• Download URL:
  https://rosap.ntl.bts.gov/view/dot/91813/dot_91813_DS1.pdf
• File Type:
  [PDF - 9.42 MB ]