Research On Gesture Recognition Based On Wifi Singal

WiFi signal is a type of radio wave in 2.4GHz and 5.8GHz band,which is widely used in short distance wireless communication and large data transmission due to its many advantages such as short wavelength,high frequency and large bandwidth.With the development of pattern recognition and human-computer interface technology,the powerful capability of WiFi signals in target detection and recognition is gradually discovered.Now,utilizing WiFi signals,researchers have already realized target localization,human activity and gesture recognition.And such technology is attracting more and more interests.This paper not only discusses some significant aspects of WiFi signal based gesture recognition technology including signal processing,feature extraction and classification,but also proposes an effective gesture recognition model.When Wi-Fi signals pass through moving hands,their propagation characteristics will be affected and the way they are influenced is determined by the hand' moving patterns,which means the signals is to some extent modulated by gestures and contain information about the motions.By processing and analyzing the received Wi-Fi signals,we are able to extract the information and realize recognition.Generally,to realize gesture recognition,a gesture model has to be established first by conducting proper data collection and preprocessing procedures.Then,feature extraction algorithms should be applied to extract features from the established model.And a classifier is also required for classification of extracted features.At last,the effectiveness of proposed recognition model must be proved by conducting tests.In this paper,the data collection procedure is implemented on Sora,a software defined radio platform.The long preamble of received OFDM frame is collected as raw data.Then,a power profile is acquired by preprocessing raw data.And a similarity-based segmentation method is applied to split the power profile into fragments which are considered as gesture models.In order to decrease the data dimension and eliminate noise,the discrete wavelet transform is applied to extract features.And a support vector machine(SVM)improved by dynamic time warping(DTW)algorithm is built to classify different gestures.The experiment result shows that our method can recognize nine predefined dynamic gestures with an average recognition rate of 94.8%using only a small amount of training samples,which proves its effectiveness and potential for practical applications.