Asilla, Inc. (Headquarters: Machida, Tokyo; CEO: Go Onoue; hereinafter “Asilla”), whose vision is “a safe and comfortable world through the power of technology,” conducted a proof of concept on preventing accidents at level crossings using behavior recognition AI, jointly with Odakyu Electric Railway Co., Ltd. (Headquarters: Shinjuku-ku, Tokyo; President: Shigeru Suzuki; hereinafter “Odakyu Electric Railway”). The two companies have announced that they presented the results as a co-authored paper at the 63rd Railway Cybernetics Symposium, held on Thursday, November 5 and Friday, November 6, 2026.
Safety measures at level crossings have improved year by year, including the introduction of obstacle detection devices. However, the most common cause of level crossing accidents is “crossing immediately before a train,” in which pedestrians or vehicles enter a crossing even though a train is approaching.*1 Such behavior is difficult to detect in advance with conventional obstacle detection devices, and new means of preventing accidents before they happen have been needed.
Odakyu Electric Railway has introduced a “centralized level crossing monitoring system” at all 228 level crossings on its lines, allowing conditions inside crossings to be checked via IP cameras. Since FY2024, the two companies have conducted a two-phase proof of concept on a system that uses Asilla’s behavior recognition AI to analyze footage from these existing cameras and detect behaviors that may signal an accident. The results have now been compiled into a paper and presented at the 63rd Railway Cybernetics Symposium, which brings together railway engineers.
Dates: Thursday, November 5 – Friday, November 6, 2026
Venue: Hotel Metropolitan Ikebukuro (Toshima-ku, Tokyo)
Organizer: Japan Railway Cybernetics Council, a special committee of the Japan Railway Engineers’ Association
Paper: “AI-Based Detection of Pedestrian Behavior at Level Crossings: Real-Time Analysis of Camera Footage Using Behavior Recognition AI”
Presenter: Odakyu Electric Railway representative (Q&A: Tatsuya Nakamura, Project Manager, Solution Development Department, Asilla)
This proof of concept targeted five behaviors that could be signs of a level crossing accident: prolonged lingering in front of a crossing, folding one’s arms after lingering, sitting down, continued lingering after a train has passed, and loitering around the crossing. The main results are as follows.
- Prolonged lingering in front of a crossing was detected with high accuracy in environments where a person could be continuously captured
- The configuration was improved to always use a mechanism that reconnects tracking interrupted by obstructions or overlapping people based on appearance features, enabling continuous tracking of the same person
- The person re-identification logic before and after a train passes was improved, raising re-identification accuracy from about 60% to about 90%
*Based on a survey by Odakyu Electric Railway and Asilla (evaluation results from the proof of concept conducted in FY2024–FY2025)
The verification also confirmed that shooting conditions, such as camera position, angle of view, and sunlight, affect detection performance. Asilla will continue to improve its detection logic, aiming to help prevent level crossing accidents in a wider range of environments.
The two companies will proceed with selecting level crossings for deployment, aiming to use the technology to prevent accidents at level crossings.
Leveraging the knowledge gained through its work with Odakyu Electric Railway, Asilla will contribute to solving the challenges faced by railway operators, such as ensuring safety and maintenance work.
Representative: Go Onoue, CEO
Address: 1-4-2 Nakamachi, Machida-shi, Tokyo, Japan
Business: Development and provision of products and solutions based on behavior recognition AI
Website: https://en.asilla.com/
Asilla, Inc.
PR: Nakamura
Email: pr@asilla.jp
.png)