Detection of Intoxicated Passengers at Stations to Prevent Accidents on Railway Platforms

Accidents due to falls from station platforms occur frequently in Japan, with more than half caused by intoxicated passengers. However, effective preventive measures, such as the installation of platform doors, are expensive and difficult to implement in small and medium-sized stations. As a new preventive measure, this study proposes a low-cost wobble detection system that requires only a camera and a lightweight computer. In the proposed method, wobbliness is detected by evaluating the pattern of fluctuations in the skeleton coordinates over a specific time frame. We conducted various experiments to evaluate the proposed wobble detection approach in simulated station situations. By assessing the presence or absence of wobble at three locations (hip, knee, and ankle), we found the accuracy rates to be 95.0% for the hip, 93.3% for the knee, and 91.7% for the ankle, demonstrating high accuracy.