Research On Location-independent Gesture Recognition Based On WiFi

With the rapid development of wireless communication technology,people find that wireless signals have the ability to recognize human gestures within the sensing range.The WiFi-based gesture recognition technology,as an important part of smart home,has attracted more and more attention in this field due to its advantages of no additional equipment and no invasion of privacy.The basic principle of this technology is that the propagation path of the WiFi signal is affected by the movement of the human body,so that the reflected signal exhibits distinct characteristics related to different activities,and the Doppler Frequency Shift(DFS)can reflect this.Therefore,gesture recognition can be performed by establishing the connection between DFS and gesture actions.However,due to the fact that WiFi signals can sensitively perceive changes in the surrounding environment in practice,the development and application of gesture recognition technology based on WiFi signals is partially limited,therefore,when the user’s location and other environmental factors change significantly,it is difficult for the model trained on a specific location to achieve ideal results on data from other locations.So how to achieve location-independent perception in practical applications is still a challenge.This paper provides a position-independent gesture recognition method based on WiFi signal according to the problem that the recognition accuracy of the model is reduced due to the position change when the user performs gesture actions.The main contents of this paper are as follows:(1)In this article,we analyze the corresponding changes of WiFi signals caused by various gestures made by the human body in the Fresnel area in the indoor environment,and finally performs denoising and phase processing on the Channel State Information(CSI),The first two principal components are obtained by principal component analysis(PCA)to increase the features of the input,and the time-frequency analysis method is used to visualize the DFS as the feature input of the neural network model.According to the theory of wavelet transform,In this article,we design a gesture recognition model based on wavelet convolutional network,and replaces the first layer of convolutional convolutional network(Convolutional Neural Network,CNN)with a wavelet convolutional layer.The final experimental results show that under the action of wavelet convolution layer,the recognition accuracy of this method on public datasets and self-collected datasets can reach 94.3% and90%,espectively.(2)In order to solve the position-dependent problem of WiFi signals,In this article,we design a model structure based on transfer learning.By fixing the parameters of the convolutional network and updating the parameters of the fully connected layer,the position-independent gesture recognition can be realized.task,and use data augmentation to improve the overall gesture recognition accuracy.The experimental results show that the recognition accuracy of the method can reach 89.3% on the new position without training.