Development of an unsupervised 3D LiDAR-based methodology for automated safety monitoring of railway facilities

Railway safety (e.g., at grade crossings, platforms, or rail tracks) is a primary concern for transportation authorities. Unfortunately, preventable railway collisions claim the lives of hundreds annually, often involving individuals crossing illegally at highway-railway grade crossings or trespassing at unauthorized railroad facilities. Transportation authorities often deploy a range of engineering countermeasures to mitigate the frequency or risk of such events. These countermeasures include technological solutions that automatically activate warning systems, barriers, or gates to alert and deter road users from unlawfully entering restricted railway facilities. For the safety monitoring of such facilities, alternative sensing technologies such as video-based computer-vision systems have been evaluated and, in some cases, utilized in practice. Despite their merits, implementing automated LiDAR-based detection and tracking methods has yet to be explored in railway safety applications. This research aims to introduce and assess an unsupervised 3D-LiDAR-based methodology for monitoring rail-road level facilities. This study’s core is the implementation of an unsupervised learning algorithm designed to detect, track, and classify road users using point clouds gathered by a 3D-LiDAR sensor. The proposed methodology demonstrates encouraging results when monitoring rail-road level crossings. The aggregate average absolute percentage deviation (AAPD) for motorized road users and counting motorized road users stands at 5% and 3%, for non-motorized road users at 10% and 14% on two separate test days, each featuring distinct system installations.