sEMG-Based Continuous Hand Action Prediction by Using Key State Transition and Model Pruning.

Area of Science:

• Biomedical Engineering
• Robotics
• Human-Computer Interaction

Background:

• Surface electromyography (sEMG) is widely used for hand motion classification and joint angle estimation.
• Current methods for discrete motion recognition have poor real-time performance.
• Continuous hand action prediction based on motion continuity remains an underexplored area.

Purpose of the Study:

• To develop a robust method for continuous hand action prediction.
• To improve real-time performance in gesture recognition.
• To enhance the reusability of predictive models for multigesture actions.

Main Methods:

• Proposed a key state transition model using Gaussian Mixture Models-Hidden Markov Models (GMM-HMMs) as a condition for prediction.
• Implemented a sliding window with long-term memory to dynamically detect key state transitions.
• Utilized model pruning to enhance the reusability of models for continuous multigesture prediction.

Main Results:

• Achieved an average accuracy of 97% for continuous two-hand actions with a 70 ms time delay.
• Outperformed Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models in accuracy and time delay.
• Attained over 85% accuracy with a 90 ms delay for continuous four-hand actions.

Conclusions:

• The proposed key state transition method offers superior real-time performance for continuous hand action prediction.
• This approach significantly advances the field of gesture recognition and human-computer interaction.
• The method demonstrates potential for real-world applications requiring precise and rapid motion interpretation.