Widar3.0: Zero-Effort Cross-Domain Gesture Recognition With Wi-Fi.

Area of Science:

• Computer Science
• Signal Processing
• Human-Computer Interaction

Background:

• Wi-Fi devices offer opportunities for human gesture recognition via wireless signals.
• Current Wi-Fi gesture recognition systems require domain-specific adaptation or re-training for new environments.
• Existing methods necessitate extra data collection or model re-training when encountering new data domains.

Purpose of the Study:

• To develop a Wi-Fi-based zero-effort cross-domain gesture recognition system named Widar3.0.
• To enable gesture recognition across different domains without requiring explicit adaptation or re-training.
• To achieve fully zero-effort recognition by addressing the limitations of current domain-specific approaches.

Main Methods:

• Proposing Widar3.0, a novel Wi-Fi-based gesture recognition system.
• Extracting domain-independent features from wireless signals at a lower signal level.
• Developing a one-fits-all general model trained once for cross-domain adaptation.

Main Results:

• Achieved 92.7% accuracy for in-domain gesture recognition.
• Demonstrated cross-domain recognition accuracy ranging from 82.6% to 92.4% without model re-training.
• Outperformed state-of-the-art solutions in cross-domain gesture recognition tasks.

Conclusions:

• Widar3.0 successfully achieves zero-effort cross-domain gesture recognition using Wi-Fi signals.
• The system's ability to extract domain-independent features allows a single model to adapt to diverse environments.
• This approach significantly advances the practical usability of Wi-Fi-based gesture recognition systems.