Real-Time Analysis of Hand Gesture Recognition with Temporal Convolutional Networks

Area of Science:

• Biomedical Engineering
• Machine Learning
• Signal Processing

Background:

• Electromyography (EMG)-based hand gesture recognition is crucial in biomedical engineering.
• Deep Learning, particularly Convolutional Neural Networks (CNNs), is widely used for this task.
• Temporal Convolutional Networks (TCNs), a class of CNNs, have shown recent success.

Purpose of the Study:

• To evaluate the real-time performance of TCNs for EMG-based hand gesture recognition.
• To assess a TCN model's suitability for sequence classification in this domain.
• To identify limitations of TCNs in real-time applications.

Main Methods:

• Electromyography-based hand gesture recognition framed as a sequence classification problem.
• Utilized Temporal Convolutional Networks (TCNs).
• Conducted simulation experiments on a recorded surface EMG (sEMG) dataset to evaluate real-time behavior.

Main Results:

• A proposed TCN network with data augmentation showed a marginal accuracy improvement over existing models in offline training.
• Real-time evaluation demonstrated a decrease in classification accuracy.
• The TCN architecture was found unsuitable for real-time EMG-based gesture recognition applications.

Conclusions:

• Real-time analysis is essential for understanding model limitations in practical scenarios.
• The study highlights the need for further research to improve TCN performance for real-time gesture recognition.
• Investigating alternative approaches or model modifications is necessary for future development.