IMG2Speed: Generative AI and Multimodal Machine Learning for Predicting Operating Speed Distributions from Roadway Design and Context

Designers set target speeds to achieve safe operations, yet observed operating speeds often diverge because the influence of geometric and contextual elements (e.g., lane width, medians, trees, curb extensions, etc.) is not quantified in a way that is practical for design. A modern data-driven machine learning approach can be a potential solution to learn the quantitative mapping from observable design elements to operating speed distributions. This project proposes to (i) automate data curation from spot-speed reports using Generative Artificial Intelligence (AI) like Large/Vision Language Models (LLMs/VLMs); (ii) fuse the curated evidence base with street-view imagery and Geographic Information System (GIS)/context layers to extract geometric and streetscape attributes; and (iii) develop a machine learning (ML) model that estimate the percentiles of operating speeds used in practice (e.g, median, 85th) from cross-section and visual/context features.

• Record URL:
  • https://researchprojects.dot.state.mn.us/projectpages/details?id=31994&type=CONTRACT

Language

• English

Project

• Status: Active
• Funding: $166,847.00
• Contract Numbers:

  1058110 WO#11

• Sponsor Organizations:

  Minnesota Department of Transportation

  395 John Ireland Boulevard
  St Paul, MN  United States  55155
• Managing Organizations:

  Minnesota Department of Transportation

  Office of Research & Innovation
  395 John Ireland Boulevard, MS 330
  St. Paul, MN  United States  55155-1899
• Performing Organizations:

  University of Minnesota, Minneapolis

  Minneapolis, MN  United States  55455
• Principal Investigators:

  Choi, Seongjin

• Start Date: 20260701
• Expected Completion Date: 20280630
• Actual Completion Date: 0

Subject/Index Terms

• TRT Terms: Artificial intelligence; Data analysis; Design speed; Highway design; Machine learning; Operating speed; Speed distribution
• Subject Areas: Data and Information Technology; Design; Highways; Operations and Traffic Management; Planning and Forecasting;

Filing Info

• Accession Number: 01995043
• Record Type: Research project
• Source Agency: Minnesota Department of Transportation
• Contract Numbers: 1058110 WO#11
• Files: RIP, STATEDOT
• Created Date: Jul 8 2026 4:24PM