The U.S. railroad system is comprised of approximately 830 railroads and 210,000 grade crossings (FRA, 2018). Railroads have been continuously addressing the issue of trespassing at highway-rail grade crossings that may potentially result in serious consequences. In 2018, 263 deaths and 840 injuries occurred at highway-rail grade crossings in the United States. Most previous trespassing-related studies were based on publicly available accident data. However, the trespassing-related accident data only represents a small proportion of all trespassing events, the greater portion of which being trespassing events. The increasing deployment of cameras in railroad systems can contribute to the collection of trespassing data, but it is very labor intensive to review the video data manually. To leverage the untapped potential of the big video data for railroad trespassing risk management, this study develops an Artificial Intelligence-aided trespassing detection technique and has processed around 1,700 hours of video data from a stationary camera installed at one grade crossing in New Jersey. In this case study, over 3,000 highway-rail grade crossing violations were identified by our developed Al algorithm, and recorded in a trespassing database. The distributions of trespassing events by hour of the day, day of the week, daylight period, trespasser type (e.g., pedestrian, car, truck, bus, motorcycle, etc.), and other risk factors are analyzed and presented. Finally, potential mitigation solutions are proposed from engineering, enforcement, and education perspectives.
展开▼