Wi-Fi-Based Location-Independent Human Activity Recognition via Meta Learning.

Area of Science:

• Computer Science
• Artificial Intelligence
• Signal Processing

Background:

• Wi-Fi-based device-free human activity recognition is crucial for IoT and HCI applications.
• Current methods struggle with location variations and require extensive data, hindering real-world deployment.
• Location-independent sensing is essential but challenging due to data limitations and positional changes.

Purpose of the Study:

• To develop a location-independent human activity recognition system using Wi-Fi signals.
• To address the challenge of limited datasets and adverse effects of location variations on recognition accuracy.
• To enhance the generalization and transferability of human activity recognition models.

Main Methods:

• Utilized a Convolutional Neural Network and Long Short-Term Memory (CNN-LSTM) for location-independent feature extraction.
• Proposed a metric learning-based approach for activity recognition to improve model transferability across different locations with limited data.
• Conducted extensive experiments in an office environment across 24 testing locations.

Main Results:

• Achieved over 90% accuracy in location-independent human activity recognition.
• Demonstrated effective adaptation to limited data samples, including a small number of subcarriers and low sampling rates.
• Significantly promoted the generalization and transferable capability of the recognition model.

Conclusions:

• WiLiMetaSensing offers a robust solution for location-independent human activity recognition using Wi-Fi.
• The proposed system effectively mitigates the impact of location variations and data scarcity.
• The approach shows promise for practical applications in IoT and HCI, even under resource-constrained conditions.