The number of incidents, fatalities, and injuries occurring at highway-rail grade crossings (i.e., intersections where a road and railway line are at the same grade or level crossings for short) has remained in the same range within the past 15 years. According to the Federal Railroad Association (FRA), there were 2,192 collisions, 247 fatalities, and 766 injured at all public and private crossings in 2023 [1]. Additionally, there were 47 suicide casualties resulting in 42 fatalities and 5 injuries [2]. Safety provisions at level crossings has largely gone unchanged for decades. Active crossings have devices for controlling traffic such as flashing lights, gates, audible alarms, etc. Passive crossings do not have any traffic control devices and road users must use their own judgement to determine when it is safe to cross. While smart technology at road environments has provided benefits, smart technology at level crossings seems to be lagging typical road environments like intersections. This project proposes to establish a digital and hardware infrastructure testbed with V2X communication to enable a research, development, and deployment platform for computational methods and alarming systems for railway crossings. We have partnered with the National Robotics Engineering Center to deploy the testbed on a structure overlooking at least one level crossing. See supplemental information for prospective location of testbed where two railway tracks create two crossings with the road. The high vantage point of the testbed will provide an unoccluded birds eye view of the crossing in order to provide information fast enough for the train/vehicle to respond. The testbed will consist of two cameras, one facing each direction along the railroad tracks, hardware for V2X communication, and a computer for collecting and storing data. For this project, we will develop computational methods for 1) detecting and localizing trains, vehicles, and VRUs, 2) assessing a ‘danger’ score for all detections, and 3) estimating distance and time to intersection for trains. Through this development, we will collect a novel dataset of visual images/videos and train a novel AI model. We will demonstrate these methods by wirelessly relaying alarming information to a wireless device, e.g., smartphone. The platform will also be able to receive communications from trains in order to send alarming information to the level crossing for vehicles and VRUs. In order to develop this component of the platform, we will develop methods, collecting data, and training AI models to detect vehicles and VRUs at level crossings. [1] https://oli.org/track-statistics/collisions-casualties-year [2] https://safetydata.fra.dot.gov/officeofsafety/publicsite/Query/suiabbr.aspx