Background: Muscle fatigue of the lower limbs results in dynamic imbalance and gait instability, increasing the risk of falling. However, people might slow walk without physical muscle fatigue due to mental fatigue. Wearable inertial measurement units (IMU) and machine learning approaches have been well employed for recognizing human activities.
Research question: The study aims to use a machine learning technique to recognize the data collected from IMUs for physically fatigued or slow-walking gaits. Second, the study aims to reveal the location or the number of IMUs can have the best performance.
Methods: Sixteen healthy adults with six IMUs attached to their heels, toes, sacrum, and head participated in the experiment. On the first day, the participants were instructed to walk along a hallway before and after the fatigue protocol as the Pre- and Post-fatigue gait. On the second day, the participants were instructed to walk along a hallway following the beat of their fatigue gait cadence measured on the first day as the simulated cadence (SC) gait. Gait cycles of each condition were segmented as the inputs of the Long Short-Term Memory (LSTM) model for recognization.
Results: The result revealed that the LSTM model could recognize the gait of simulated cadence with the highest accuracy among these three gaits. For the signal body part, the highest accuracy was 93.20 % observed at the IMUs of toes. For the best combination, the IMUs of toes and sacrum achieved the highest accuracy of 95.71 %.
Significance: The machine learning technique of LSTM with one or more IMUs can recognize the gait under normal, physical fatigue, or simulated cadence without muscle fatigue. Our model and approach would be expected to provide conditional warning in multiple fields, such as industrial safety for potential applications.
Keywords: Gait recognition; Inertial measurement units; Long short-term memory; Muscle fatigue.
Copyright © 2022 Elsevier B.V. All rights reserved.