Train Data Collection and Arrival Time Prediction System for Highway–Rail Grade Crossings

Abstract

A key component of the safe and efficient operation of traffic signals near highway–rail grade crossings (HRGCs) is an accurate estimate of a given train's arrival time at each crossing. An improvement in the accuracy and timing of the predicted train arrival would allow for improvements in traffic signal preemption algorithms, which would, in turn, lead to increased driver and pedestrian safety and reduced network delay. This paper introduces a system for train data collection and arrival time prediction at an HRGC that uses real-time data from a combined video–Doppler radar vehicle detection system. Train speed data were collected at a dual-track test bed system in Lincoln, Nebraska. Both kinematic equation–based and multiple linear regression models were developed and used to predict train arrival time at the HRGC. The best models, according to average absolute errors (AAE), were identified. On average, regression models were more accurate than kinematic models by approximately 13%. As detection time increased, AAEs of both the kinematic and regression models decreased. In addition, the confidence intervals about mean prediction errors were obtained with a bootstrap method. The average prediction errors of the regression models were approximately 50% smaller than the kinematic models, all else being equal.