A Deep Learning Hybrid Model for Identifying Gait Patterns and Transition States of Lower Limb Exoskeleton Wearer

Abstract

Addressing the synchronization issue between lower limb exoskeletons and the limbs of wearer, this article proposes a deep learning hybrid model for identifying gait patterns and transition states using only lower limb angle data. The proposed hybrid model integrates 1-D convolutional networks with BiGRU. The model accurately predicts subsequent gait patterns by identifying transition states, thus achieving smoother transitions during movement and enhancing wearer comfort. Compared to other models, this article introduces a novel evaluation index named HIAM, which demonstrates the comprehensive performance advantage of the proposed model. The model classifies five gait patterns and eight transition states using only the angle data from the thighs and shanks of wearer. The classification accuracy and ${F}1$ -score are 99.05% on the validation set, achieving 99.11% accuracy and ${F}1$ -score on the test set. The HIAM reaches 99.29 on the validation set and 99.36 on the test set.