Lidar could improve collision avoidance for trains

Lidar tends to be associated with autonomous vehicles, given that to everyone except Elon Musk, it’s considered an essential tool in helping AVs  “see.”

But lidar—which stands for light detection and ranging—can put in work beyond AVs, helping to adjudicate Fat Bear Week, assist archeological digs, and—via a recent partnership—allow trains to detect obstacles at railway crossings.

South Korea-based Seoul Robotics is behind the latter application, partnering with Herzog Technologies in late March to develop the Critical Asset Monitoring (CAM) system, an obstacle-detection service for railway crossings. Herzog operates its own train collision avoidance systems in over 4,500 trains throughout the United States.

Since its founding in 2017, Seoul Robotics has sold its lidar-perception software to industries including robotics and autonomous vehicles, and cities looking to implement smart-city solutions, and raised $18.1 million in funding.

Spawned out of an idea to use lidar to test the perceptive range of a single sensor, the CAM system uses lidar sensors to observe and pinpoint when cars, trucks, buses, and people encroach into a railway intersection, alerting train operators in real time to slow down and avoid major, potentially fatal accidents. According to the US Department of Transportation, there have been 19,342 railway collisions, resulting in 2,260 fatalities, since 2013.

“As long as you have a physical shape, as long as you’re not a ghost, the system is able to detect whether something is there or not,” HanBin Lee, CEO of Seoul Robotics, told Emerging Tech Brew. “First example, cameras, you have to train those…to identify certain objects, or the object has to move in order for the camera system to understand. For us, as long as it has a physical shape, we’ll process it.”

That's because lidar “sees” by emitting pulses of light and then receiving them back, like sonar or radar, but with lasers instead of sound or radio waves. Those pulses can then be used to quickly create precise 3D models of the environment around the sensor.

It can take the average train traveling 55 mph an entire mile—or the length of ~18 football fields—to stop, making earlier warning systems crucial. Lee said the distance at which train operators are warned depends on the circumstances, but said the operator “could be miles away.”

Of course, the system isn’t perfect. Lee said it had some early issues with picking up and classifying certain objects.

“The first time the system was deployed, the train was classified as a bus, because the field of view and the angle and the length of the train within the field of view was almost identical,” Lee said. “We didn’t know there was a lot of dust in Texas. So when a train or bus soars by, it leaves a trail of dust. We don’t have that in Seoul. We don’t have a lot of dust…Those are the edge cases.”

Lee said CAM-equipped railway crossings normally use at least two sensors on either side of the crossing to cover the entire intersection, which allows data from one sensor to cover anything missing from the first, eliminating any potential blind spots. In some cases, the tech is also used in trainyards to help workers move and back up trains without the use of mirrors.

Zoom out: So far the CAM system is in its early stages—Lee said it’s finished testing and is being deployed to “multiple sites” with private train operators in South Korea, Texas, and California. Seoul and Herzog have collaborated with Trinity Railway Express, a commuter rail operator between Fort Worth and Dallas, Texas, to use the technology.