Research On Gesture Recognition Technology Based On Millimeter Wave Radar

Gesture recognition is an important research direction in the new generation of Human Computer Interaction(HCI)technology,which has a wide range of application prospects in automotive collision avoidance,medical and entertainment.Traditional gesture recognition methods are limited to the influence of acquisition equipment,site and other environmental factors,and there is a risk of privacy leakage.The radar-based gesture recognition technology is not affected by the site and environmental factors and easy to place and other advantages.In this paper,the millimeter wave radar-based gesture recognition technology is studied as follows.1.Millimeter wave radar gesture recognition system design and gesture database establishment.Seven common gesture actions in daily life are defined,and each gesture action has its practical meaning.The radar parameters are designed according to the defined gesture actions,the gesture information is collected,the gesture action database is established,and the data set is pre-processed using the moving target display algorithm to remove redundant information.2.Feature extraction of gesture echo signal,analysis of radar IF signal,construction of gesture distance-time map and Doppler-time map,and normalization of single feature map,fusion of two feature spectra using weighted average fusion method to obtain feature fusion spectra,and extraction of micro-Doppler features of gesture actions.3.The gesture recognition classification is studied,and the residual network model is proposed to recognize and classify the gesture feature atlas.In the recognition classification experiments,two single feature maps were tested for recognition classification,and the results showed that the two single feature maps were able to achieve more than 85% recognition;after that,the feature fusion spectrograms were classified and recognized,and their accuracy was as high as 98.58%,which was 6%-7% higher than that of the single feature maps,indicating that the feature fusion spectrograms were able to represent gesture actions completely.To verify the advantages of the feature fusion spectrogram,the micro-Doppler feature map was tested for classification,and the recognition accuracy was obtained as 95.74%,which decreased compared with the recognition accuracy of the feature fusion spectrogram,indicating that the feature fusion spectrogram has outstanding advantages in gesture recognition classification.In addition,to verify the relationship between the number of residual network layers,four network models,Res Net18,Res Net50,Res Net101 and VGG16,were used for recognition and classification,and the results showed that the recognition accuracy of the Res Net101 network model was the highest,indicating that the network model can achieve better recognition and classification with the increase of the number of network layers.The millimeter wave radar-based gesture recognition system proposed in this paper achieves the recognition and classification of seven predefined gestures with good recognition accuracy of98.58%.