Keystroke Recognition using WiFi Sensing: An Empirical Study on Robustness

Abstract

With the widespread use of WiFi networks, they have become a valuable tool for detecting physical environment information. Recent studies have revealed that WiFi Channel State Information (CSI) can be utilized to detect physical events such as movement, occupancy increases, and gestures. Our research aims to determine the effectiveness, reliability, and readiness of WiFi sensing technology for detecting finger keyboard strokes. We conducted experiments to predict how accurately WiFi CSI can detect keys pressed on a keyboard. To start with, our experiments showed that in an ideal setting with a robotic arm, the position of a specific key on the keyboard can be predicted with 99% accuracy. We also successfully achieved human finger localization over a key and actual key- press recognition, with reduced accuracies of 94% and 89%, respectively. Our empirical findings testing the robustness suggest that WiFi sensing technology is strong, at least in static and controlled indoor environments. However, our experiments also highlight the limitations of WiFi sensing technology in small- scale gesture recognition for generic settings, emphasizing the importance of device positioning in maximising the underlying technology's efficacy.