Citation

Chang J, Walker CL. Prevent. Treat. Nat. Disasters 2022; 1(2): e65.

Copyright

(Copyright © 2022, Scientific Publishing)

DOI

10.54963/ptnd.v1i2.65

PMID

unavailable

Abstract

Millions of motor vehicle crashes and tens of thousands of resulting deaths occur each year in the United States. While it is well known that wintry conditions make driving more difficult and dangerous, it is difficult to quantify and communicate the threat to motorists, especially in real-time. This proof-of-concept research uses machine learning (ML) to approach this problem in a new way by creating a ML model that can identify snow on the road from a traffic camera image. This information is coupled with the number of coincident vehicular crashes to provide detailed consideration of the impact of snow on the road to motorists and transportation agency decision-makers. It was found that, during meteorological winter, when the ML model determined there to be snow on the road in a traffic camera image, the chance of a vehicular crash pairing with that traffic camera increased by 61%. The systems developed as part of this research have potential to assist roadway officials in assessing risk in real time and making informed decisions about snow removal and road closures. Moreover, the implementation of in-vehicle weather hazard information could promote driver safety and allow motorists to adjust their driving behavior and travel decision making as well.

Copyright (c) 2022 Joshua Chang, Curtis Louis Walker

Language: en

Keywords

image classification; image recognition; machine learning; road condition; road weather; weather-related crashes