Wi-SL: Contactless Fine-Grained Gesture Recognition Uses Channel State Information

Area of Science:

• Human-Computer Interaction
• Wireless Sensing Technology
• Biomedical Engineering

Background:

• Gesture recognition is crucial for human-computer interaction, with sign language being a key application.
• Existing methods often require users to wear devices, limiting practical usability.
• Effective sign language recognition can significantly aid individuals with aphasia and hearing impairments.

Purpose of the Study:

• To propose a novel contactless, fine-grained gesture recognition method for sign language using WiFi Channel State Information (CSI).
• To develop a system that enhances interaction for individuals with communication disabilities.
• To achieve high-precision sign language recognition without requiring any wearable devices.

Main Methods:

• Utilized commercial WiFi devices to map wireless signal subcarrier amplitude and phase differences to sign language actions.
• Implemented an efficient denoising technique and optimal subcarrier selection to reduce system computational cost.
• Employed a K-means and Bagging optimized Support Vector Machine (SVM) classification (KSB) model for enhanced action data classification.

Main Results:

• The proposed Wi-SL method achieved an average accuracy of 95.8% in gesture recognition across three different scenarios.
• Demonstrated the effectiveness of device-free, non-invasive sign language recognition.
• Validated the system's capability for fine-grained gesture recognition.

Conclusions:

• Wi-SL offers a high-precision, device-free solution for sign language recognition.
• The system has the potential to improve computer interaction and daily life for individuals with communication impairments.
• This approach leverages existing WiFi infrastructure for advanced human-computer interaction applications.