research
The Door Camera Learns to Spot the Last-Second Dash
A single-frame model predicts risky subway boarding behavior before doors close, with reported accuracy of 97.58 percent.
Summary
A single-frame model predicts risky subway boarding behavior before doors close, with reported accuracy of 97.58 percent.
Sungkyunkwan University researchers developed a computer-vision system intended to flag passengers at risk of being trapped by closing subway doors. The model uses one CCTV frame to classify risky boarding behavior and reported 97.58 percent accuracy in the study dataset, aiming to warn operators before an entrapment occurs. The result is a research evaluation, not evidence of performance across every station, camera angle, crowd, weather condition, or deployed transit network.
Why it matters
A single-frame model predicts risky subway boarding behavior before doors close, with reported accuracy of 97.58 percent.
Limits and context
- The result is a research evaluation, not evidence of performance across every station, camera angle, crowd, weather condition, or deployed transit network.
Key claims
A single-frame model predicts risky subway boarding behavior before doors close, with reported accuracy of 97.58 percent.
Qualification: The result is a research evaluation, not evidence of performance across every station, camera angle, crowd, weather condition, or deployed transit network.
Evidence: source-2026-07-21-006
Sources
- Sungkyunkwan University via Newswise: Subway door entrapment AISungkyunkwan University via Newswise · official announcement
Corrections
No corrections have been recorded for this story.