Category
Methods / Statistics
Document Type
Paper
Abstract
The purpose of this study was to automatically identify the key gait events, foot-strike and foot-off, from 2D video data. Markerless motion capture and pose estimation have become accepted tools in many biomechanics applications to automatically analyse 2D videos of human movement. However, the accurate detection of gait events from various camera views is still a challenge. We trained a long short-term memory neural network to identify foot-strike and foot-off events in walking and running trials captured from nine different camera views based on 2D pose estimation keypoint labels. We achieved a detection accuracy of 86.3-96.1% (F1 score 76.2-92.5%). These results show the applicability of machine learning tools for the automatic detection of key event frames, which will help practitioners to easily identify frames of interest for further biomechanical analyses.
Recommended Citation
Mundt, Marion; Colyer, Steffi; and Alderson, Jacqueline
(2024)
"DETERMINATION OF GAIT EVENTS FROM 2D VIDEO USING LONG SHORT-TERM MEMORY NEURAL NETWORKS,"
ISBS Proceedings Archive: Vol. 42:
Iss.
1, Article 151.
Available at:
https://commons.nmu.edu/isbs/vol42/iss1/151
