GAIT EVENT DETECTION DURING WALKING USING DEEP LEARNING AND THIGH-WORN ACCELEROMETRY

Authors

Claas Lendt, German Sport University Cologne, Auckland University of TechnologyFollow
Tom Stewart, Auckland University of TechnologyFollow

Category

Wearable Technology

Document Type

Paper

Abstract

Mobility and gait are important indicators of human health. However, measuring them outside of a laboratory setting can be challenging. To measure human physical behaviour in free-living conditions, thigh-worn accelerometers and data-driven algorithms are commonly used. This study explores a deep learning approach that utilises data from a single thigh-worn accelerometer. A temporal convolutional network was trained to predict gait events in healthy adults during various walking conditions. The model demonstrated a high level of detection accuracy (F₁ score ≥ 99%) and good time agreement for both gait events, with the 95% limits of agreement being -0.04s and 0.04s. Minor differences in spatiotemporal gait parameters were observed. The results indicate the potential of using a deep learning approach with thigh-worn accelerometry data for future research.

Recommended Citation

Lendt, Claas and Stewart, Tom (2024) "GAIT EVENT DETECTION DURING WALKING USING DEEP LEARNING AND THIGH-WORN ACCELEROMETRY," ISBS Proceedings Archive: Vol. 42: Iss. 1, Article 165.
Available at: https://commons.nmu.edu/isbs/vol42/iss1/165