Computer Vision Model Trained to Detect Passenger Fall, Slip, and Trip Events in Rail Stations Achieved an Accuracy of 82 Percent.
Video and Still Images from Rail Stations Around the World Were Analyzed To Automatically Extract and Classify High Risk Behavior.
Nationwide, United Kingdom
A Deep Learning Approach Towards Railway Safety Risk Assessment.
Summary Information
Rail stations are often busy environments that have a high risk of fall, slip, and trip (FST) events. Factors such as floor conditions, visibility, footwear, and crowding may contribute to the risk of FSTs on station stairs, platforms, and escalators. Failing to monitor stations can result in compromised passenger safety as well as negative impacts to rail operations. CCTVs and staff observation may aid in detecting risk, but computer vision also has potential to automatically identify risks by recognizing unsafe actions in real time. This study’s goal was to manage the risk of falls by detecting and analyzing passengers automatically using data from CCTV cameras.
METHODOLOGY
This study proposed a deep learning-based computer vision model to detect adverse conditions by capturing FST events. It relied on image-detection methods and introduced a risk management framework to analyze images and videos. The study gathered data from open sources from multiple countries; data availability was noted as a limitation because the images and videos are typically deleted due to privacy concerns and other reasons. The model automatically detected and classified passenger behaviors (e.g., standing in risky position, falling, full fall posture), to support decision-making for alerting station control and the train driver.
FINDINGS
- The model was effective for detecting fall, slip, and trip–related risky behaviors in railway stations, with a precision score (how many detected events are correct) ranged from .70 to .92 and a recall score (detection coverage) ranged from .75 to .90.
- The proposed method can enable real-time FST event identification, help reduce injuries and liability, improve data quality for safety management, support cost savings, and provide valuable insights for future infrastructure planning and improvements.
