Railway safety has long been a concern due to the immense size and speed of trains, coupled with their path through populated areas. The average train is more than a mile long, and at 55 mph, it requires a mile to stop when the brakes are fully deployed. Collisions can also cause train derailment and potential environmental dangers with toxic spillage. In 2023, there were 2,174 collisions in the US, according to the Federal Railroad Administration, causing 250 fatalities and 737 injuries, highlighting the urgent need for improved safety measures.
To address this issue, Rail Vision, based in Tel Aviv, in collaboration with NVIDIA Metropolis, has unveiled a groundbreaking AI vision camera that can be affixed to trains as an early detection system. This technology can detect potential obstacles, including humans and vehicles, from a distance of 1.2 miles in real-time, under various weather conditions, and day or night. Inspired by a tragic 1985 train collision near Tel Aviv, that claimed 22 lives, including 19 children in a school bus, Rail Vision CEO Shahar Hania and his team developed this early detection system, driven by a commitment to preventing future accidents.
“By joining Metropolis, Rail Vision hopes to gain a significant advantage in developing and deploying AI-driven solutions that are tailored to the unique needs of the rail sector through the opportunity of leveraging NVIDIA’s extensive technological expertise and integrating its cutting-edge technology. Rail Vision uses the NVIDIA Jetson and other NVIDIA edge AI platforms, which provide accelerated computing in compact and energy-efficient modules, and also uses the NVIDIA TensorRT software development kit for high-performance deep learning inference.”
Powered by NVIDIA’s Jetson Xavier modules, specifically designed for autonomous machines, the AI camera utilizes advanced algorithms for visual odometry, obstacle detection, and path planning. This enables the camera to provide timely visual and acoustic alerts to cargo and passenger train operators, helping them react swiftly to potential hazards. Moreover, the system can seamlessly integrate with a train’s infrastructure to automatically apply brakes when an obstacle is detected, enhancing safety even further.
The device resembles a sleek home theater projector, and can be affixed atop the front, or the face, of the train. The camera identifies the object and places a virtual box around it and tracks its movement and distance. Additionally, the camera displays “impact time” under the existing speed and distance.
The railroad industry has benefitted greatly from AI technology in the past decade, performing tasks in minutes that would take inspection crews days to perform, and with a greater degree of accuracy.
Norfolk Southern, a railroad freight company that has been in business since 1827 in the U.S., operating in 22 states, has taken advantage of AI technology in improving safety and efficiency. According to the company, it utilizes AI for autonomous track, and train, inspection and inventory.
“As our trains transport goods across the nation, our cutting-edge, car-mounted imaging systems and AI models are busy inventorying every rail — including the manufacturer, year, size and condition. We use this information to create a digital twin of our entire network. With this digital model, we can remotely analyze the condition of our rails and proactively address the safety of our tracks.”
Norfolk Southern also incorporates AI into its planning. “Building on our extensive rail expertise, we’re using big data to make track safety even smarter. Our predictive AI models evaluate the digital twin of our network to look for patterns and plan for maintenance as rail conditions change. These rail-health algorithms, powered by machine learning, make it possible to predict the need for rail maintenance up to five years in the future.”
Additionally, the company uses AI models to analyze images of their trains as they move through high-speed, high-resolution cameras that collect images of the train at different angles, including from underneath. “Our industry-leading AI models look at these images to accurately and independently identify defects — and send an alert to our mechanical wayside desk for maintenance if a defect is found.”
