Meitetsu EI Engineer Co., Ltd.

Using AI and cameras mounted on utility poles near railroad crossings, the railroad crossing monitoring system keeps railroads running smoothly and local communities safe.

Keeping Numerous Railroad Crossings Safe with AI Image Analysis Technology

Meitetsu EI Engineer’s main customer, Nagoya Railroad, currently has about 1,000 railroad crossings on all of its railway lines. One of our important missions is to keep these crossings safe because railroad crossings, where roads meet railroad tracks, can be prone to serious accidents. One way to keep railroad crossings safe is to use obstacle detection systems, but they can be difficult to install due to the diverse environments at railroad crossings. This is why not all crossings operated by Nagoya Railroad have obstacle detection systems in place. In addition, because obstacle detection systems work by emitting lasers from both ends of the crossing, they can easily detect large objects such as cars, but struggle to detect smaller objects like people. The accuracy of detecting people can be improved by using 3D LiDAR, and railway companies in the Tokyo metropolitan area have installed it. However, due to the extremely high cost of 3D LiDAR systems, Nagoya Railroad needed a more affordable system that it could install at its numerous railroad crossings.
In response to these problems, Nagoya Railroad developed a plan to improve safety at railroad crossings by installing not only status monitoring equipment for barriers, emergency stop buttons, and other devices, but also cameras. Additionally, they decided to use AI and other technologies to further improve safety beyond just monitoring the equipment. This was because people would sometimes step onto the tracks at railroad crossings in rare cases, which the old system failed to detect. With the decision to install cameras, we began to explore a system that would improve safety with AI-driven image analysis technology, similar to how 3D LiDAR systems detect unusual events.

Contec’s Edge AI Computer Met the Requirements

To meet Nagoya Railroad’s request, we consulted with several manufacturers. Toyota Systems, with its experience in AI image analysis, was eager to take on the challenge and joined the project. Toho Denki Industry, a specialist in railway equipment, also came on board, and development began in earnest.
Although the project framework was in place, we faced many problems. One of the toughest challenges was the process of training the AI, which is critical to the system. Because railroad crossings are installed in all kinds of environments, it is not always possible for the AI to consistently recognize similar situations. We fed the system with as much data as we could, and if there were false positives, we repeated the training process to improve the accuracy to the target level.
While the AI training process was underway, we combined the expertise of all the companies in the project team and held repeated discussions to establish the safety protocols, such as defining the appropriate operations if unusual events were detected and how to coordinate with the status monitoring equipment. To achieve Nagoya Railroad’s goal of reducing as many railroad crossing accidents as possible, we positioned the system as a “safety support device” rather than a full-fledged “safety device,” which must meet strict safety standards. This is to keep deployment costs low and allow it to be installed at more railroad crossings.
In addition, we developed the AI using an NVIDIA evaluation kit equipped with Jetson, so we needed to use an edge AI computer equipped with Jetson for the actual deployment. This project involved multiple companies, so to avoid complications due to differing international business practices in case any problems arise, we decided to use domestically manufactured products wherever possible. It was then that we learned from Contec that they were planning to release a new product with Jetson. We have used Contec’s products in the past to develop a different monitoring system. We think highly of their products because of their consistent and reliable performance. We also appreciate their reliable customer support framework, so we chose them again without hesitation. It also helps that their products support digital I/O, which is necessary to connect to our various systems. We chose the DX-U2200 because it meets our current specifications and allows for future expansion.

Railroad crossing monitoring system screen

Taking AI Image Analysis a Step Further for Manpower Saving and Improved Efficiency

This railroad crossing monitoring system is designed to detect unusual events and change traffic signals and stop trains when it determines a potential accident. It is important that the system does not cause frequent train stops due to false alarms or malfunctions. The system is currently deployed at Nagoya Railroad’s nine railroad crossings, and there have been no major issues so far. We feel that it is even more stable than when we tested it with the evaluation kit prior to its deployment.
In the future, we aim to achieve our target number of installations. We also plan to use image analysis technology to reduce manual inspections at crossings and to save manpower. In addition, we believe that tracking traffic volume at crossings at various times and analyzing this data will help improve the safety and reliability of train operations. As we add more features, we may require a different edge AI computer. We look forward to exploring Contec’s product lineup again in future.

Implementation Points

Adopted Products

• 
  DX-U2200

  • Edge AI accelerator powered by NVIDIA Jetson Orin NX
  • Built-in NVMe storage
  • 4G LTE Cat.4 communication model can be selected
  • Four Gigabit LAN ports for alternating connections between different network layers
  • -20 to 60°C ambient temperature

Return to Implementation Case Studies