Recognizing intentions with body segmental cues of gait cycles before direction change during continuous walking
1Department of Industrial Engineering and Engineering Management, National Tsing Hua University, Hsinchu, Taiwan.
Background:
Gait intention is typically detected using electroencephalogram (EEG) and primarily focuses on recognizing the initiation of walking. Recently, wearable sensors have been extensively used to classify different walking patterns.
Research Questions:
The study was aimed to investigate the effectiveness of Inertial Measurement Units (IMUs) in recognizing directional change intentions during gait, hypothesizing that distinct patterns exist that indicate the initiation of turning before actual turning movements occur.
Methods:
Twenty healthy participants performed one to two gait cycles before directional changes during straight walking, a 45-degree right turn, and a 90-degree right turn. Seven IMUs were attached to various body segments, and a matching network combined with sliding window bidirectional gated recurrent units was used to recognize turning intentions. The EEG data analysis was used by EEGLAB to determine the event-related desynchronization in the alpha (8-13 Hz) and beta (14-30 Hz) frequency bands.
Results:
The results showed that the model using different combinations of IMU sensors achieved an optimal accuracy of 96.80 %, with several combinations exceeding 95 %. Subtle movements in different body segments were sufficient to predict upcoming turning intentions. The accuracy dropped to 83.79 % in the model that excluded data from the head segment, while the other models that included the head segment achieved over 94 % accuracy. Furthermore, angular acceleration data was notably more accurate, at 93.46 %, compared to 88.43 % for acceleration data alone.
Significance:
Successfully highlighted the crucial role of sensing contralateral body and head segments, providing insights into the model's interpretability. The use of IMUs for gait turning intention recognition shows promise in replacing devices that rely on sensing brain waves. Furthermore, this approach opens possibilities for applications such as controlling exoskeletons, including assisting walking robots with strategically placed sensors.
