Analysing driver behaviour and crash frequency at railway level crossings using connected vehicle and GIS data

Railway level crossings (RLCs) pose unique safety challenges as crucial intersections between vehicular traffic and railways. This study examines the effects of driver behavior on the safety performance of RLCs (within a 150-meter radius) in New South Wales, Australia. Historical databases on crashes, train operations, and inventory for RLCs were integrated. Also, vehicle movement raw data, including acceleration, deceleration, G-force, and speed, were extracted using the connected vehicle data. Then, the driver’s harsh braking and stiff steering events were identified. A random effect Hurdle Poisson model was adopted to account for the excessive zeros and unobserved heterogeneity. We identified risk factors affecting the likelihood and number of crashes at RLCs across two severity levels. Results of the non-injury crashes suggested that street, active control types, harsh braking, and stiff steering were associated with a higher likelihood of zero crash observations, while train frequency showed the opposite effect. It was also revealed that multiple tracks contributed to the increase in the number of non-injury crashes. Interestingly, after crossing the hurdle, harsh braking at RLC is associated with more non-injury crashes. As for the injury crashes, active control and stiff steering were associated with a higher likelihood of observing zeros. Nevertheless, once crossed the hurdle, intersecting with a highway, having multiple rail tracks and harsh braking events show increasing effects on the frequency of injury crashes. The findings of this study emphasize the need for targeted driver education programs and a larger-scale implementation of active control measures. It is hoped that these actionable insights can assist policymakers in prioritizing safety interventions at RLCs.