Design And Implementation Of Letter Input System Based On Gesture Recognition Using WiFi Signal

With the development of computer technology,HCI(Human-Computer Interaction)is becoming more and more important in people's daily life.As a category of gesture recognition technology,writing recognition has received much attention from researchers for its simpleness,universality,and convenience.Traditional gesture recognition technology can be classified into three categories:Camera-Based,Sensor-Based,Hardware-Based.Camera-Based technology is limited to Line-of-Sight and brightness,and may cause privacy issues.Sensor-Based technology requires the user to carry sensor device,which sometimes is inconvenient for people.Hardware-Based technology utilizes expensive specialized device,which costs much and is difficult to be deployed.To address these problems of traditional gesture recognition,we design a handwritten capital recognition system using CSI(Channel State Information)called Wi-Wri.CSI can be extracted from commercial WiFi devices and contains PHY channel state information which provides many fine-gained multi-path propagation characteristics.The main work of Wi-Wri can be divided into two parts:The first part is detection and extraction for characteristic waveform based on CSI.We first leverage Butterworth and PCA(Principal Component Analysis)to remove noises of CSI data.Then we propose a writing detection algorithm to detect and extract the characteristic waveform caused by writing.The second part is matching and recognition for waveform feature based on CSI.We first leverage DWT(Discrete Wavelet Transform)to reduce the number of the CSI data.Then we use DTW(Dynamic Time Warping)as the tool to obtain minimum distance alignment between any two waveforms and get the recognition result leveraging kNN(k-Nearest Neighbor).Finally,we proposed CEC(Context-based Error Correction)and further improve recognition accuracy.Experimental results show that Wi-Wri can achieve 98.1% detection accuracy and 82.7%recognition accuracy in indoor scenarios.