Deepfine, Seoul Metro to develop AI tram safety platform
Field testing will begin on the Wi-Rye Line in the first year, with detection accuracy targeted above 95 percent.
Deepfine announced on October 20 that it will develop an AI-based safety operation platform for trams together with Seoul Metro. The company is the lead research institute on the project.
The research is being carried out under the 2026 AX Innovation Enterprise Creative Technology Development Program, run by the Ministry of Science and ICT and the Institute for Information & Communications Technology Planning & Evaluation (IITP). The project began in April 2026 and is scheduled to run until December 2027, including follow-up research with stage evaluations.
Deepfine will develop multi-sensor risk detection technology, 3D spatial modeling and a digital twin control system, and smart glasses-based driver support technology. Seoul Metro will define risk types and operating standards and take part in field testing on the Wi-Rye Line.
In the first year, AI technology will fuse data from CCTV and LiDAR to detect risks around the tracks. Detected risk is classified into three levels — safe, caution and danger — with metrics including TTC (Time to Collision). The first-year field testing will be conducted on the Transit Mall section between Wi-Rye Lake Park and Wi-Rye Station Park.
Target performance goals are over 95% detection accuracy, display of warnings on the control screen and smart glasses within 0.5 seconds, and over 90% capture of objects in non-visible areas.
Quick answers
What is Deepfine developing with Seoul Metro?
An AI-based safety operation platform for trams, covering multi-sensor risk detection, 3D spatial modeling, a digital twin control system and smart glasses-based driver support.
Where will the field testing take place?
On the Wi-Rye Line, in the Transit Mall section between Wi-Rye Lake Park and Wi-Rye Station Park during the first year.
What are the target performance goals?
Over 95% detection accuracy, warnings on the control screen and smart glasses within 0.5 seconds, and over 90% capture of objects in non-visible areas.