Discrimination Models of Swing Phase and Stance Phase using MediaPipe

Abstract

It is useful to extract the time factor of gait from captured gait videos. However, this requires frame-by-frame counting, which is expensive in terms of manpower and time. In this study, we developed discrimination models to discriminate between the stance phase and the swing phase of a walker from a walking video. Using MediaPipe, a marker-less motion capture system that can be used with a single camera, we discriminated the stance phase and swing phases of a walker from the angular changes of the waist, knee, and ankle on each side of the walker in each frame. The results showed that the right leg and left leg were discriminated with 95.1% and 95.0% accuracy, respectively. The gait cycle was calculated from the discrimination results, and the average deviation was only 7.4% for the right leg and 4.2% for the left leg.