Indoor Personnel Detection And Posture Recognition Based On Millimeter Wave Radar

Traditional human target detection and recognition technologies,such as optical imaging and image processing,are increasingly showing limitations.Radar detection of human targets has become a research hotspot.Millimeter wave radar has the advantages of penetration,non-contact,non-disclosure of privacy,immunity from light intensity,and all-weather applications.It can be widely used in smart homes,medical monitoring,life detection,and disaster relief.This paper mainly studies the detection of personnel based on millimeter wave radar and posture recognition based on radar micro-Doppler signatures.Personnel detection includes the detection of human vital signs and identification based on respiratory signals.Millimeter wave radar IWR1443 is used to collect vital signs signals(mainly respiratory and heartbeat signals)of static human bodies and posture echo data of dynamic human bodies.Signal processing and analysis are conducted on the measured data,and neural networks are mainly used in the classification stage.The main research contents are as follows:(1)The transmission,reception and IF signal model of FMCW radar are described,the positioning principle of FMCW radar is analyzed,and the parameters such as range,speed and angle are deduced.The hardware structure of FMCW radar is described,and the analysis process of received data of MIMO TDM radar is also analyzed.(2)The detection of human vital signs based on phase signals is studied.In the research of single target vital sign detection,range FFT and phase unwrapping are used to extract vital sign phase signals from radar echo data.Breathing and heartbeat signals are separated using a method based on bandpass filtering and modal decomposition.The Pearson correlation coefficient PCC is used to improve the variational modal decomposition(VMD)and optimize the selection of the modal decomposition number K value.In terms of respiratory and heartbeat frequency estimation,two spectral estimation algorithms,MVDR and PMTM,are proposed and compared with traditional FFT.In the research of multi target vital sign detection,based on range FFT,azimuth estimation algorithms for angle FFT,MUSIC,and Capon are proposed.High resolution localization of multiple targets is achieved through comparison,and the phase signals of each target are extracted separately.(3)The identification based on respiratory signals is studied.The respiratory timing signal obtained by signal processing is used as the input of a single type network LSTM,which only extracts the temporal characteristics of the respiratory signal,and the resulting identification rate needs to be improved.An improved LSTM hybrid model based on 1D-CNN is proposed.1D-CNN extracts the spatial features of respiratory signals through convolutional layers,and then inputs a sequence of spatial features into LSTM to extract temporal features.Experimental results show that the hybrid model can effectively improve the recognition accuracy of five identities,with an average accuracy rate of 93.8%,significantly improving the classification effect compared to a single network classification.(4)The recognition of human posture and behavior based on radar micro-Doppler signatures is studied.Simulate the Boulic human body model,and compare and analyze the radar echo of the Boulic human body model under walking posture in time and frequency domain through various time-frequency transformation methods.Six different human posture behaviors are measured using millimeter wave radar,and a micro-Doppler signatures dataset of human posture is constructed through signal processing such as STFT.In the classification stage,an improved residual network based on APRe LU-ST is proposed,which simplifies the network and improves the recognition accuracy to 91.7%.