Enhancing Safety Analysis Through Predictive Modelling at Highway Railroad Crossing

Abstract: Abstract: Safety at highway-railroad grade crossings (HRGCs) is of great importance because of its catastrophic consequences. To enhance traffic safety and reduce accidents neat HRGCs, it is essential to accurately predict vehicle trajectories and identify possible near-miss collisions. In this paper, the Britton HRGC in Oklahoma City, one of the most dangerous interstates in the state, was selected as the case study. Field video streaming data were collected in the field and used to detect vehicles in each of the approach, forecast their future trajectories, and detect potential crash locations Real-time trajectory data are extracted from the implementation of YOLO V8 in combination with BYTEtrack for vehicle detection and tracking. A multilayer LSTM model was then deployed for the prediction of trajectories of vehicles are predicted. Subsequently, vehicle factors, such as vehicle speed and traveling direction, were determined to estimate precise vehicle locations and predict the time to collision (TTC) as a surrogate measure for possible vehicle collisions. Meanwhile, the patterns of risky driving behaviors were identified and then correlated to the observed historical crash data, through which the significant influencing factors to HRGC safety could be identified. This study highlights the advantages of real-time collision detection systems in enhancing safety at HRGCs. The integrated system from this study could help better assess intersection safety and identification of hotspots with informed travel behaviors.

Keywords: Highway Railroad Grade Crossing (HRGC), vehicle trajectory, Time to Collision (TTC), safety