Wrist-Worn Hand Gesture Recognition While Walking via Transfer Learning.

Area of Science:

• Human-Computer Interaction
• Biomedical Engineering
• Signal Processing

Background:

• Walking introduces movement artifacts that degrade hand gesture recognition accuracy.
• Existing methods are often validated only in static poses, ignoring dynamic effects.
• Accurate hand gesture recognition during dynamic activities like walking is crucial for real-world applications.

Purpose of the Study:

• To develop and validate a signal decomposition approach for segmenting gestures during walking.
• To implement a transfer learning method for domain adaptation between walking and standing scenarios.
• To demonstrate the feasibility of accurate hand gesture recognition using wrist-worn sensors during dynamic walking.

Main Methods:

• Empirical Mode Decomposition (EMD) was used for signal decomposition to isolate gestures.
• A transfer learning approach with distribution adaptation was applied for domain transfer.
• Experiments involved ten subjects performing seven hand gestures using an IMU wrist-worn device during walking and standing.

Main Results:

• The signal decomposition method reduced gesture detection error by 83.8%.
• The transfer learning approach improved recognition accuracy by 15.1% with a 20% transfer rate.
• The study successfully demonstrated accurate hand gesture recognition during dynamic walking.

Conclusions:

• A novel signal decomposition and transfer learning framework enables robust hand gesture recognition during walking.
• Wrist-worn sensing combined with advanced algorithms can overcome motion artifacts.
• This work paves the way for ubiquitous adoption of hand gesture recognition in dynamic human-machine interaction.