Dynamic Gesture Customization in a Continous Learning Framework: Empowering Users with Personalized Real-Time sEMG...

Area of Science:

• Human-Computer Interaction
• Biomedical Engineering
• Machine Learning

Background:

• Dynamic gesture recognition systems require continuous adaptation to new gestures.
• Existing frameworks may struggle with efficient integration of novel gestures without performance degradation.
• Adaptive continuous learning frameworks are crucial for real-world applications.

Purpose of the Study:

• To present an extended dual-model adaptive continuous learning (DM-ACL) framework for dynamic gesture addition.
• To evaluate the performance of DM-ACL using both offline and online incremental learning.
• To demonstrate the framework's capability for continuous accuracy enhancement.

Main Methods:

• Implementation of an extended dual-model adaptive continuous learning (DM-ACL) framework.
• Evaluation using offline and online incremental learning approaches for new gesture integration.
• Collection of surface electromyography (sEMG) data for gesture recognition.

Main Results:

• Online incremental learning achieved 82.69% overall accuracy, 82.8% precision, 82.69% recall, and 0.83 F1 score with three new gestures.
• The DM-ACL framework allows for continuous accuracy enhancement towards preset thresholds.
• Efficient data collection requires only 4 seconds of sEMG data per new gesture.

Conclusions:

• The extended DM-ACL framework effectively manages dynamic gesture sets without significant performance loss.
• The framework demonstrates suitability for real-world applications in healthcare and human-computer interaction.
• Continuous learning within DM-ACL enables ongoing improvement of gesture recognition accuracy.