sEMG-signal and IMU sensor-based gait sub-phase detection and prediction using a user-adaptive classifier

Area of Science:

• Robotics
• Biomedical Engineering
• Machine Learning

Background:

• Surface electromyogram (sEMG) signal analysis for gait sub-phase detection faces challenges including recognition delay and subject-specific signal variations.
• Existing methods often struggle with feature relevance and user adaptability in lower-limb robotic applications.

Purpose of the Study:

• To develop an improved gait sub-phase detection and prediction system for lower-limb power-assist robots.
• To address limitations of current sEMG-based gait analysis, focusing on accuracy, speed, and user adaptation.

Main Methods:

• A novel labeling technique based on heel and toe events was developed.
• Muscle and feature selection strategies were employed to enhance signal relevance.
• A user-adaptive classifier utilizing weighted voting and interpolation for prediction was implemented.

Main Results:

• The proposed labeling technique improved accuracy by 7% compared to existing physical sensor methods.
• Muscle and feature selection enhanced accuracy by 12%, and the user-adaptive classifier by 17%.
• The developed prediction technique demonstrated an 80% faster average prediction time.

Conclusions:

• The novel approach significantly enhances the accuracy of gait sub-phase detection in lower-limb robotics.
• The user-adaptive classifier and interpolation-based prediction offer substantial improvements in speed and performance.
• This method provides a more robust and efficient solution for sEMG-based gait analysis in assistive robotic devices.