Further Development of AI Tool for Extraction of Road Hazards from Videolog Data and LIDAR
The UNC Highway Safety Research Center (HSRC), the Renaissance Computing Institute (RENCI), and potentially the Volpe National Transportation Systems Center (Volpe) are proposing to develop the NCDOT Safety Data Initiative (SDI) Artificial Intelligence (AI) Tool, the feasibility of which has been established in the one-year project funded by the USDOT SDI State and Local Government Data Analysis Tools for Roadway Safety grant (to be completed in November 2021). In the initial implementation of the tool, the research team, in close partnership of NCDOT Traffic Safety, Operations Program Management, and Research and Development units, achieved a prediction accuracy of 90% for guardrails and 88% for the utility poles in collected video log data in 2018 and 2019 by NCDOT for all secondary roads. To further meet NCDOT needs, the research team is proposing to extend the approaches implemented in the initial version of the tool to assign geometric locations to safety-related features. The research team will further integrate LIDAR and videolog data. These two extensions will provide a greater range of features including topographical features and provide the geometric information necessary to make use of the identified features in a safety context. The proposed work will use a combination of AI, classical computer vision and imaging algorithms to add geometry to the predictions so that every point or extended object will have a real-world location extracted from the data. NCDOT has acquired aerial LIDAR data for the whole state of North Carolina. In collaboration with the NCDOT Photogrammetry unit, the research team will use NCDOT LIDAR data to compute roadway geometry derivatives, such as side slope, lane width, centerline of the road, edge of the road, density of trees, and surface curvature. In this proposed upgrade of the tool, the fusion of LIDAR with videolog data will allow combined analysis of these sources, using the best properties of each to enhance the results. With these upgrades to the NCDOT SDI tool, the research team will use the fused LIDAR/videolog dataset to extract a set of point features such as poles and mailboxes, extended features such as guardrails, and topographical elements such as side slope and cliff faces.
Language
- English
Project
- Status: Active
- Funding: $215750
-
Contract Numbers:
2023-22
-
Sponsor Organizations:
North Carolina Department of Transportation
Research and Development
1549 Mail Service Center
Raleigh, NC United States 27699-1549 -
Managing Organizations:
North Carolina Department of Transportation
Research and Development
1549 Mail Service Center
Raleigh, NC United States 27699-1549 -
Project Managers:
Bolyard, Stephanie
-
Performing Organizations:
University of North Carolina Highway Safety Research Center
730 Martin Luther King Jr. Boulevard CB # 3430
Chapel Hill, NC United States 27599-3430 -
Principal Investigators:
Srinivasan, Raghavan
- Start Date: 20220801
- Expected Completion Date: 20240731
- Actual Completion Date: 0
Subject/Index Terms
- TRT Terms: Artificial intelligence; Hazards; Highway safety; Image processing; Laser radar; Machine learning; Pattern recognition systems
- Geographic Terms: North Carolina
- Subject Areas: Data and Information Technology; Highways; Safety and Human Factors;
Filing Info
- Accession Number: 01857542
- Record Type: Research project
- Source Agency: North Carolina Department of Transportation
- Contract Numbers: 2023-22
- Files: RIP, STATEDOT
- Created Date: Sep 13 2022 12:04PM