Winter Safety Improvement with Computer Vision and Transfer Learning

The preservation of road safety in snowy regions during the winter season is of paramount significance due to the presence of perilous meteorological circumstances, such as snowstorms, which can result in escalated vehicular collisions and subsequent roadway closure. In the present investigation, the primary objective is to devise a novel methodology aimed at tackling the aforementioned obstacle. This is achieved through the utilization of a hybridized system that incorporates both thermal and optical imagery to identify snow accumulation on road surfaces. By employing transfer learning techniques in conjunction with the U-Net architecture implemented in the Keras framework, the authors' approach demonstrates notable efficacy in attaining precise outcomes, even when confronted with the limitations imposed by a restricted dataset. The results demonstrate notable mean pixel accuracy (MPA) scores of 88% for roadway snow detection based on optical images captured during daytime and 94% based on thermal images acquired during nighttime. The encouraging results observed in this study underscore the potential of dual-spectrum imaging technique to greatly improve road safety and reduce the number of collisions in winter conditions.

• Record URL:
  • https://www.ugpti.org/resources/reports/details.php?id=1240
• Record URL:
  • https://rosap.ntl.bts.gov/view/dot/79664
• Summary URL:
  • https://rosap.ntl.bts.gov/view/dot/79665
• Supplemental Notes:
  • This document was sponsored by the U.S. Department of Transportation, University Transportation Centers Program.
• Corporate Authors:

  University of Utah, Salt Lake City

  Department of Civil and Environmental Engineering
  Salt Lake City, UT  United States 

  Mountain-Plains Consortium

  North Dakota State University
  Fargo, ND  United States  58108

  Office of the Assistant Secretary for Research and Technology

  University Transportation Centers Program
  Department of Transportation
  Washington, DC  United States  20590
• Authors:
  • Ramezanpourkami, Moein
  • Zhu, Xuan
• Publication Date: 2024-10

Language

• English

Media Info

• Media Type: Digital/other
• Edition: Final Report
• Features: Figures; Maps; Photos; References; Tables;
• Pagination: 22p

Subject/Index Terms

• TRT Terms: Data analysis; Highway safety; Image analysis; Machine vision; Snow; Thermal imagery
• Subject Areas: Data and Information Technology; Highways; Safety and Human Factors;

Filing Info

• Accession Number: 01946164
• Record Type: Publication
• Report/Paper Numbers: MPC-698, MPC 24-572
• Files: UTC, NTL, TRIS, USDOT
• Created Date: Feb 18 2025 10:45AM