Related Experiment Video
Updated: Sep 29, 2025
08:15
Capturing Dynamic Finger Gesturing with High-resolution Surface Electromyography and Computer Vision
Published on: March 28, 2025
789
WiFi Signal-Based Gesture Recognition Using Federated Parameter-Matched Aggregation
Weidong Zhang1,2, Zexing Wang1,2, Xuangou Wu1,2
1School of Computer Science and Technology, Anhui University of Technology, Maanshan 243032, China.
Sensors (Basel, Switzerland)
|March 26, 2022
Summary
This study introduces WiMA, a WiFi signal-based gesture recognition system. WiMA enhances model robustness and accuracy using matched averaging federated learning for diverse environments.
Area of Science:
- Computer Science
- Electrical Engineering
- Artificial Intelligence
Background:
- WiFi signal-based gesture recognition is crucial for smart homes.
- Existing methods struggle with environmental variations and require extensive data.
- Centralized training models are accurate but lack robustness and require many participants.
Purpose of the Study:
- To develop a robust WiFi signal-based gesture recognition system.
- To address the limitations of traditional federated learning in heterogeneous environments.
- To improve gesture recognition accuracy without large centralized datasets.
Main Methods:
- Proposed a novel system named WiMA (WiFi-based gesture recognition with matched averaging federated learning).
- Utilized neuron arrangement invariance in parameter aggregation for federated learning.
- Implemented a distributed gesture recognition environment with seven participants.
Main Results:
- Achieved an average accuracy of 90.4% in gesture recognition.
- Demonstrated improved robustness against heterogeneous Channel State Information (CSI) data.
- Performance closely matched state-of-the-art centralized training models.
Conclusions:
- WiMA offers a robust and accurate solution for WiFi signal-based gesture recognition.
- The matched averaging federated learning approach effectively handles environmental variations.
- This system reduces reliance on large, centralized datasets and participant numbers.
