Related Experiment Video
Updated: Jun 26, 2025
08:15
Capturing Dynamic Finger Gesturing with High-resolution Surface Electromyography and Computer Vision
Published on: March 28, 2025
455
Machine learning for hand pose classification from phasic and tonic EMG signals during bimanual activities in virtual
Cédric Simar1, Martin Colot1, Ana-Maria Cebolla2
1Machine Learning Group, Computer Science Department, Université Libre de Bruxelles, Brussels, Belgium.
Frontiers in Neuroscience
|May 13, 2024
Summary
This study introduces a new virtual reality system for recording EMG signals and hand movements, improving myoelectric prosthesis control. Integrating physiological knowledge enhances machine learning models for better gesture recognition and prosthesis function.
Area of Science:
- Biomedical Engineering
- Neuroscience
- Machine Learning
Background:
- Advances in machine learning and bioelectrical signal acquisition are improving myoelectric prostheses for upper limb loss.
- Current methods for decoding electromyography (EMG) signals face challenges in accurately interpreting hand gestures.
Purpose of the Study:
- To introduce and validate a novel experimental paradigm for synchronized EMG recording and hand pose estimation using virtual reality.
- To compare different hand gesture classification pipelines using EMG signals.
- To investigate the impact of integrating neurophysiological knowledge into machine learning models for enhanced myoelectric prosthesis control.
Main Methods:
- Developed a virtual reality headset with hand-tracking for synchronized EMG and hand pose data acquisition.
- Compared classification pipelines including standard signal processing, convolutional neural networks, and Riemannian geometry on EMG data.
- Integrated neurophysiological 'move command' information to improve EMG signal component separation.
Main Results:
- Demonstrated the effectiveness of Riemannian geometry on raw or xDAWN-filtered EMG signals for gesture classification.
- Showed significant improvement in sustained posture recognition by integrating physiological knowledge ('move command') into machine learning models.
- Validated the novel VR paradigm for EMG and hand pose data acquisition.
Conclusions:
- The proposed VR paradigm effectively captures synchronized EMG and hand pose data.
- Riemannian geometry offers a robust approach for EMG-based gesture classification.
- Integrating neurophysiological insights into machine learning models significantly enhances myoelectric prosthesis control, paving the way for more natural gesture decoding.
