AI Predicts Subway Door Accidents with 97% Accuracy Using Single Camera Frame

July 21, 2026 - 10:02

A new artificial intelligence system promises to prevent subway door entrapment accidents by predicting risky passenger behavior before it happens. The technology analyzes a single frame from existing CCTV cameras and achieves 97.58% accuracy in identifying passengers who might get caught in closing doors.

The system works by detecting subtle body language and movement patterns that indicate a passenger is about to rush through closing doors, attempt to board after the warning chime, or squeeze into an already crowded car. Unlike previous systems that required multiple video frames or complex sensor arrays, this AI needs only one image to make its prediction in real time.

Subway door entrapment accidents remain a serious safety issue worldwide. Passengers get caught by bags, clothing, or body parts when they try to board at the last second. These incidents can cause injuries, delays, and even fatalities. Current prevention methods rely on door sensors that detect physical obstructions, but these only activate after contact has already been made.

The new predictive approach shifts the safety paradigm from reactive to proactive. By warning station staff or triggering an automatic door hold before the passenger reaches the door, the system could prevent accidents entirely. The AI processes video feeds from existing security cameras, meaning no new hardware installation is needed for most subway systems.

Researchers trained the model on thousands of hours of station footage, teaching it to distinguish between normal boarding behavior and high-risk actions. The system can recognize when a passenger is running, carrying bulky items, or hesitating at the door line. It also factors in crowd density and door opening duration.

Subway authorities in several major cities have expressed interest in testing the system. If deployed widely, the technology could reduce platform accidents by a significant margin while maintaining smooth train operations. The next phase of development focuses on reducing false alarms and integrating the AI with existing train control systems.