Gait Phase Classification via Attention-Based Deep Learning withShank-Mounted IMU Signals on Level Ground and Stairs

Citations

SCOPUS

0

초록

Gait patterns vary significantly depending on terrain characteristics, highlighting the need for analytical approaches capable of capturing signal differences that arise across diverse walking environments. In this study, we propose a classification-based deep learning gait phase detection model utilizing acceleration and angular velocity data collected from shank-mounted inertial measurement units (IMUs) in three terrain conditions: level-ground walking, stair ascent and stair descent. The proposed model employs a Convolutional Neural Network–Bidirectional Hybrid Long Short-Term Memory/Gated Recurrent Unit with Attention architecture to effectively capture spatial features, temporal dynamics, and salient motion characteristics in an end-to-end manner. The dataset consists of level-ground walking data from 35 healthy adults and stair ascent/descent data from 10 healthy adults. According to the experimental results, the proposed model achieves an micro-averaged F1-score of 87.87% for both Level walking and stair walking tasks. These findings confirm that an end-to-end deep learning framework can provide practically applicable gait phase classification performance using only minimal shank-mounted IMU data.

키워드

Deep Learning; End-to-End Learning; Gait Phase Detection; Inertial Measurement Unit
제목
Gait Phase Classification via Attention-Based Deep Learning withShank-Mounted IMU Signals on Level Ground and Stairs
저자
Choi, Wonseok; Kim, Seung-Jong; Choi, Mun-Taek
DOI
10.1145/3793928.3793936
발행일
2026
유형
Conference Paper
저널명
Proceedings of 2026 10th International Conference on Control Engineering and Artificial Intelligence, CCEAI 2026
페이지
33 ~ 38