Multi-feature gait recognition with DNN based on sEMG signals.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Gait recognition using surface electromyography (sEMG) signals offers a promising biometric modality.
• Existing methods face challenges in stability and accuracy, particularly with lower limb sEMG data.

Purpose of the Study:

• To develop and evaluate a novel gait recognition method utilizing deep neural networks (DNNs) for improved accuracy and stability.
• To compare the performance of the proposed DNN-based method against traditional classifiers like Support Vector Machine (SVM) and Extreme Learning Machine (ELM).

Main Methods:

• Extraction of time-domain features (mean absolute value, root mean square, waveform length, zero-crossing points) and frequency-domain features (mean power frequency, median frequency) from lower limb sEMG signals.
• Combination of extracted time and frequency domain features into a multi-feature representation.
• Training a deep neural network (DNN) classifier for gait recognition and comparing its performance with SVM and ELM.

Main Results:

• The DNN-based gait recognition method demonstrated a superior recognition rate compared to SVM and ELM.
• The proposed DNN method achieved an average recognition rate exceeding 95% across participants.
• Low standard deviation values (0.46% to 0.94%) between subjects indicate high robustness and stability of the method.

Conclusions:

• Deep neural networks provide a highly accurate and stable approach for gait recognition using lower limb sEMG signals.
• The multi-feature combination strategy enhances the discriminative power of sEMG signals for gait analysis.
• The proposed method holds significant potential for applications requiring reliable human identification and activity monitoring.