Wi-Fringe: Leveraging Text Semantics in WiFi CSI-Based Device-Free Named Gesture Recognition.

Area of Science:

• Computer Science
• Human-Computer Interaction
• Signal Processing

Background:

• WiFi-based activity recognition systems face challenges due to insufficient training data.
• Existing systems struggle with recognizing named gestures without extensive examples.
• Device-free gesture recognition is a growing area in human-computer interaction.

Purpose of the Study:

• To propose Wi-Fringe, a novel WiFi Channel State Information (CSI)-based system for recognizing named human gestures.
• To address the data scarcity problem in WiFi-based activity recognition by leveraging semantic information from text.
• To enable gesture recognition with minimal or zero training examples per activity.

Main Methods:

• Utilizing state-of-the-art semantic representations of English words (e.g., word-to-vector) and verb attributes.
• Developing a cross-domain knowledge transfer algorithm between radio frequency (RF) signals and text data.
• Collecting and evaluating data from four volunteers across diverse environments (apartment, office) for 20 named activities.

Main Results:

• Demonstrated the enhancement of named gesture recognition by incorporating semantic word representations.
• Showcased the system's ability to recognize activities with limited or no prior training examples.
• Quantified the trade-off between recognition accuracy and the number of unseen activities.

Conclusions:

• Wi-Fringe effectively recognizes named gestures using WiFi CSI, significantly reducing the need for extensive training data.
• The proposed cross-domain knowledge transfer approach bridges the gap between RF data and textual semantics.
• This system offers a practical solution for developing more robust and data-efficient WiFi-based activity recognition applications.