Citation

Zhang Z, Trivedi C, Liu X. Safety Sci. 2018; 110B: 276-285.

Copyright

(Copyright © 2018, Elsevier Publishing)

DOI

10.1016/j.ssci.2017.11.023

PMID

unavailable

Abstract

Grade-crossing trespasses are one of the greatest sources of injuries and fatalities on railways. While there is a wealth of data regarding grade-crossing accidents, near misses (or precursor events) associated with unsafe trespassing on railroad tracks are not reported, and therefore a comprehensive dataset is unavailable. This paper presents a Computer Vision (CV) algorithm to automatically detect trespassing near misses based on surveillance video footage of railway-road grade crossings. The CV algorithm is designed to be robust under changing lighting conditions over the course of the day-night cycle and works well under varying weather conditions. The algorithm is currently implemented based on data from one grade crossing in New Jersey. With minimal configuration changes, the algorithm can be adapted to various other grade crossings. Ultimately, the CV methodology can support data-driven grade-crossing near-miss risk analysis and contribute to proactive safety improvements at grade crossings.

Language: en

Keywords

Artificial intelligence; Grade crossing; Near miss; Safety; Video analytics