Abstract Manager

Although rare, collisions involving trains with motor-vehicles, pedestrians or other road users are usually fatal, and involve significant levels of trauma. Traditionally, the safety risk of an at-grade road/rail intersection has been assessed using the cross product, and other geometric features of the rail crossing.  However, given the rare and random nature of crashes at at-grade rail crossings, it is difficult to apply predictive statistical tools to help prioritize or even evaluate safety upgrades at rail crossings. Automatic Incident Detection and Monitoring System (AIDMS) is based upon the fundamental principle that there are many more incidents or near-miss crashes than there are actual crashes, and that by using the incidents as a surrogate for actual crashes, there is the potential to accumulate the data required to make evidence-based decisions regarding safety for at-grade rail crossings. Th current research project involved the development of an AIDMS that would automatically capture and report incidents (non-compliant movements) at railway grade crossings using artificial intelligence (AI) through machine-learning techniques and high-definition video. This project is the second phase of developing the AIDMS, that aims to strengthen the “proof of concept” (POC) of the AIDMS that was initiated in the Phase 1 study completed in 2019, prepared for Transport Canada’s Rail Safety Improvement Program. Two locations were selected for this research project.  Both were located in the City of Calgary: one involving a heavy rail and the other a light rail.

Preliminary results display a high level of performance, with 83% of precision and 99% of accuracy in detecting non-compliant behavior scenarios in the general case. The system performs better and worse for specific cases (specific road users, movements, times of day) and specific non-compliant behavior scenarios. Through the successful deployment of AIDMS, the research project represents a significant stride towards transforming Canada’s rail crossings into models of futuristic, data-driven safety management. In an ongoing project, the purpose of Phase 3 is to provide a cross jurisdictional standard dataset, that can be used to establish safety benchmarks for grade crossings assessments.  This data is critical for evidence-based decision-making and has the potential to inform on high value infrastructure investments, such as grade separation, and high-frequency rail corridors.