Real-time Fleet Composition via Machine Vision and AI for use in Pedestrian Safety and Risk Exposure Studies

In 2024, Georgia Tech researchers developed automated procedures to capture very consistent vehicle images using portable high-resolution video cameras positioned on Interstate overpasses. The team collected and processed more than one-million vehicle images from four locations in the Atlanta Metro Area for a State Road and Tollway Administration research project, in which a large subset of vehicle images were coded by make-and-model. The team subsequently developed machine-vision models and AI tools to identify vehicle make-and-model combinations as part of a 2024 CHEM research project, for use in a variety of pedestrian safety and risk exposure studies. In the initial model development work, the team worked in Georgia Tech’s PACE distributed computing system. However, the resulting models are so fast, the team has concluded that the AI fleet composition models can run in real-time. In this follow-on project, the research team will refine the current models to further reduce computational requirements so that the AI models can be used in edge-computing, which will process vehicle fleet composition on site, without transmitting video data to a data center. The team’s second challenge is to design and package an efficient portable computing system with a high-end graphics card that can operate under year-round temperature and humidity conditions. The team will balance system performance with power- draw and heating/cooling requirements. Overpass video is very consistent, providing elevated and unobscured rear views of vehicles. In the third phase of the project, the team will develop protocols for collecting video from major arterials and will develop machine-vision models from a wider variety of camera views (as constrained by intersection design and safe placement of equipment). The team anticipates that arterial corridor implementation will be much more complicated and that strict camera placement protocols may be needed to reach the accuracy of overpass-image-derived models. The team anticipates that equipment development (downsizing, enclosure design, heat dissipation, power consideration, etc.) and machine vision model implementation may lead to patentable inventions or licensable software. If successful equipment deployments are afforded patent protection, the team will work with Tech’s commercialization office (commercialization.gatech.edu) to develop license agreements for the manufacture of equipment and deployment of portable edge-computing systems and/or will create a GT Create-X business startup. If the USPTO rejects the patent claims, the team will release equipment specifications, software code, and technology transfer reports under open-source licensing that will allow state DOTs and their consultants to implement the systems.

Language

• English

Project

• Status: Active
• Funding: $135,000.00
• Contract Numbers:

  69A3552348329

• Sponsor Organizations:

  Office of the Assistant Secretary for Research and Technology

  University Transportation Centers Program
  Department of Transportation
  Washington, DC  United States  20590
• Managing Organizations:

  Center for Efficient Mobility

  1111 Rellis Parkway
  Bryan, Texas  United States  77807
• Project Managers:

  Ocon, Monica

• Performing Organizations:

  Georgia Institute of Technology, Atlanta

  790 Atlantic Drive
  Atlanta, GA  United States  30332-0355
• Principal Investigators:

  Guensler, Randall

• Start Date: 20260201
• Expected Completion Date: 20270514
• Actual Completion Date: 0
• USDOT Program: University Transportation Centers
• Source Data: 03-09-GT

Subject/Index Terms

• TRT Terms: Machine learning; Mobile applications; Navigational aids; Pedestrian safety; Risk assessment; Routes and routing; Traffic volume; Wayfinding
• Geographic Terms: Atlanta (Georgia)
• Subject Areas: Data and Information Technology; Operations and Traffic Management; Pedestrians and Bicyclists; Planning and Forecasting; Safety and Human Factors;

Filing Info

• Accession Number: 01979472
• Record Type: Research project
• Source Agency: Center for Advancing Research in Transportation Emissions, Energy, and Health (CARTEEH)
• Contract Numbers: 69A3552348329
• Files: UTC, RIP
• Created Date: Feb 15 2026 4:27PM