Research on Lower Limb Step Speed Recognition Method Based on Electromyography.

Area of Science:

• Robotics and Human-Computer Interaction
• Biomedical Engineering and Rehabilitation Technology

Background:

• Wearable exoskeletons require accurate human intention recognition for effective task completion.
• Traditional control signals (pressure, angle, acceleration) lack predictive capabilities for motion.
• Electromyography (EMG) signals precede movement, offering potential for prediction.

Purpose of the Study:

• To develop an advanced control system for wearable exoskeletons.
• To enhance the accuracy of recognizing human movement intention, specifically gait speed.
• To improve the predictive capabilities of exoskeleton control systems.

Main Methods:

• Fusion of a BP neural network (for generalization) and a hidden Markov chain (for timing properties).
• Utilizing electromyography (EMG) signals as the primary input for intention prediction.
• Developing a fusion discriminant model for enhanced recognition accuracy.

Main Results:

• The fused BP neural network and hidden Markov chain model achieved a recognition accuracy of 95.1%.
• This surpasses the 91% accuracy of a standalone three-layer BP neural network using identical training data.
• Demonstrates superior performance in recognizing target step speed for wearable exoskeletons.

Conclusions:

• The fusion of BP neural networks and hidden Markov chains offers a robust solution for wearable exoskeleton intention recognition.
• This approach significantly improves the accuracy of predicting gait speed and movement intention.
• The developed model enhances the capabilities of wearable exoskeletons in rehabilitation and assistive applications.