Study On Identity Recognition Technology Based On Millimeter-wave Radar

Identity recognition technology has a wide range of applications in daily life,such as computer unlocking,mobile phone payment and system login to verify your identity all the time.Traditional identification methods mainly include fingerprint recognition,face recognition,password recognition and so on.Fingerprint identification requires direct contact between human fingers and the sensor,which is a potential health hazard.Face recognition needs to collect each person's face information,there are hidden risks of privacy disclosure;Password recognition requires people to remember their own passwords rather than using the body's own characteristics,there is a risk of forgetting.This thesis puts forward a series of based on FMCW Millimeter Wave Radar identity recognition technology.By analyzing the human radar echo signal and using the corresponding deep learning classifier to identify the individual identity.The technology allows for contactless,biometric and privacy-free identification.The work includes:1.This thesis studies the theoretical basis of FMCW millimeter wave radar,and explains how to obtain the distance,velocity,Angle and other information of the object through the original echo signal of FMCW millimeter wave radar.The radar target detection was studied,and the background subtraction method and Constant False-Alarm Rate(CFAR)were compared.2.This thesis studies and proposes an identity recognition technology based on single palm/head radar echo signal,which includes target detection,static detection,in vivo detection,data preprocessing,model reasoning,multi-Chirp joint judgment and other steps.The thesis presents the preprocessing method of radar echo data normalization of PNA and the deep learning classification algorithm based on 1D-CNN,and the average recognition accuracy of which is 97.3% in the data set of 21 individuals with a total of 120,000 samples.3.This thesis studies and proposes a gesture recognition technology based on multi-feature fusion,which uses deep learning convolutional neural network to extract the features of distance Doppler diagram and distance Angle diagram,and at the same time uses some features extracted manually to assist judgment.On the basis of gesture recognition,an authentication method based on self-coding single classifier is proposed.Gesture recognition can achieve more than 95% recognition accuracy in both the open data set and the data collected by ourselves,and gesture authentication can achieve 96.7% recognition accuracy.