Wi-Wri: Fine-Grained Writing Recognition Using Wi-Fi Signals

Abstract

Writing recognition has various applications in daily life, e.g. smart homes, HCI (Human-Computer Interaction), and game world. Some existing technologies, including sensor-based, hardware-based and camera-based, suffer from various drawbacks, which prevent them to be implemented in a low cost and universal way. Recent progresses in the area of device-free sensing with WiFi signal enlighten the opportunity to address these limitations. However, state-of-the-art works show many obstacles and challenges before realization. To this end, we present Wi-Wri, a WiFi signals based device-free written letters recognition system. Specifically, Wi-Wri consists of two commercial off-the-shelf (COTS) WiFi devices, namely, a TP-LINK TL-WR842N router as the transmitter and an Intel NUC as the receiver. It leverages the Waveform of Channel State Information (CSI) values obtaining from COTS WiFi devices as the indicator of motion of hands and fingers. As our major contributions, we propose a written detection algorithm for extracting the written activity from CSI signal, a small scale motion recognition framework for rough letter input and a dictionary-based rectification to improve the input accuracy. The experimental results based on our implementation show that Wi-Wri can achieve more than 98.1% detection accuracy for detecting the motion of written letters and more than 82.7% recognition accuracy for recognizing written letters.