Research On Gesture Recognition Algorithm Based On Information Fusion Of WiFi And Millimeter Wave Radar

Gesture recognition technology has important applications in the field of humancomputer interaction.Gesture recognition technology based on WiFi signal can accurately recognize gestures in non-line-of-sight environments,but because wireless signals are sensitive to channel changes and the indoor background environment often changes,the anti-interference ability of gesture recognition is poor.Gesture recognition technology based on millimeter-wave radar signals has strong anti-interference ability,but due to the short wavelength of millimeter-wave and weak penetration ability,when there are obstacles,the recognition accuracy of millimeter-wave radar is greatly reduced.To further promote the development and practical application of gesture recognition technology,and rationally use the existing gesture recognition technology to improve the robustness of the gesture recognition system,the following research work is carried out in the thesis:First of all,this thesis uses discrete wavelet transform to denoise the original data of Channel State Information(CSI)obtained from WiFi signals,and then uses principal component analysis to reduce the dimensionality of the data;The echo data is subjected to 2D-FFT to obtain a range-Doppler map.It is proposed to superimpose multiple frames of gesture data to form a Compressed Range-Doppler(CRD)map.In this thesis,the CSI dimension reduction data and CRD map are used as the training input of the neural network,which can effectively retain the gesture feature information,reduce the input sample dimension,and improve the network training efficiency.Then,this thesis implements gesture recognition based on the WiFi platform and gesture recognition based on the millimeter-wave radar platform respectively.The data collected and processed by the two platforms are trained under classical convolutional neural networks such as Alex Net,VGGNet,and Res Net,respectively,to realize the classification of 6 kinds of dynamic gestures.According to the characteristics of the gesture recognition technologies of the two hardware platforms in different environments,this thesis proposes a gesture recognition method based on the fusion of WiFi and millimeter-wave radar information to improve the performance of gesture recognition in different environments.The experimental analysis of the information fusion method shows that the recognition performance of the dual-platform information fusion gesture recognition method in the non-line-of-sight environment and the anti-interference environment is significantly improved compared with the single platform.Finally,based on the basic network of WiFi and millimeter-wave radar information fusion,this thesis proposes a gesture recognition algorithm GAWR-FNN based on the fusion of WiFi and millimeter-wave radar information according to the characteristics of data collected from different platforms.GAWR-FNN uses a gate recurrent unit(GRU)to extract the temporal features of CSI data and uses a convolutional block attention module(CBAM)to extract the deep features of the CRD map.Through experimental analysis,it is concluded that compared with the gesture recognition method using the classical neural network as the feature extraction module for data fusion,GAWR-FNN has greatly improved the overall performance of gesture recognition,making the comprehensive accuracy rate of gesture recognition reach 98.5%.The recognition accuracy in non-lineof-sight environments and interference environments has reached 97%.