Research On Indoor Human Sensing Method Based On Millimeter Wave Radar

Indoor human sensing technology is an important bridge for the close connection and interaction between human society,information space and physical world.By acquiring comprehensive and accurate sensing information of human-objectenvironment in indoor space,indoor human sensing technology provides digital and intelligent services for applications such as smart home,health monitoring,smart aging,and human-computer interaction.Due to the advantages of passive sensing,high spatial resolution,privacy protection,and independent of weather and light,millimeter wave radar is gradually becoming a research hotspot for indoor human sensing.In this dissertation,we analyze the sensing mechanism of millimeter wave radar in indoor human sensing,and carry out the research on human gait recognition,human gesture recognition,human heartbeat detection and human Heart Rate Variability(HRV)monitoring based on millimeter wave radar for the characteristics of human sensing signal.The research has solved the problems of strong noise interference,weak sensing generalization ability,difficulty in sensing weak signals and limited accuracy of complex scenes,and enhanced the sensing capability of millimeter wave radar-based indoor human sensing technology.The main research contents are as follows:(1)To address the problems of strong clutter interference and incomplete extraction of 3D gait point cloud features in indoor human gait recognition process,this dissertation proposes a multi-person gait recognition method based on spatio-temporal feature fusion.Based on the principle of millimeter wave radar indoor human perception,a static clutter removal method based on Moving Target Indicator(MTI)and a spatial point cloud clustering noise reduction method based on DBSCAN are designed to eliminate static and dynamic clutter interference in indoor environment.On this basis,a neural network model based on spatio-temporal feature fusion is constructed to comprehensively extract and fuse the temporal and spatial information contained in the 3D gait point cloud,and finally achieve accurate multi-person gait recognition.(2)To address the problems that existing millimeter wave radar-based gesture feature extraction models are susceptible to changes in angle and distance,unable to characterize the inherent features of gestures,and weak model generalization ability,a cross-domain gesture recognition method based on a dual-stream feature model is proposed.By studying the mechanism of 3D point cloud perception of millimeter-wave radar,the characteristics of point cloud changes triggered by gesture movements are revealed.Based on the above characteristics,the spatial alignment mechanism of the millimeter wave radar 3D point cloud is designed to eliminate the perception error caused by the relative position change of the sensing target and the device.And on the basis of spatial position alignment,a cross-domain perception model based on dualstream features is constructed to finally realize accurate cross-domain perception of gesture motion.(3)To address the problems of weak heartbeat signal and serious perception difficulties caused by respiration and its harmonic interference,a heartbeat detection method based on wavelet packet transform is proposed.By establishing the fine motion model of the chest displacement caused by respiration and heartbeat motion,the intrinsic regular characteristics of the chest displacement caused by heartbeat motion are revealed.Based on the above patterns,a heartbeat signal enhancement method based on differential enhancement is designed to enhance the perception of heartbeat motion.Based on the heartbeat signal enhancement,a wavelet packet transform-based heartbeat second harmonic signal reconstruction strategy is established to suppress the serious interference of respiration and its higher harmonics on the heartbeat signal and effectively improve the accuracy of heartbeat perception.(4)To address the problems faced in HRV monitoring such as serious interference from Random Body Movement(RBM)and difficulties in heartbeat signal separation in complex scenes,this dissertation proposes a high-precision HRV monitoring method based on Variational Mode Decomposition(VMD)in complex scenes.By analyzing the thoracic motion model caused by heartbeat and respiration,an entropy thresholdbased RBM interference filter is designed to separate the interference.Based on this,a signal separator based on VMD is proposed to improve the separation effect of multiple signals and finally achieve high precision HRV monitoring in complex scenes.In summary,this dissertation investigates the millimeter wave radar-based indoor human sensing method,proposes a robust and universal millimeter wave radar indoor human sensing method,and conducts a large number of experiments in different indoor scenarios using commercial millimeter wave radar devices,and the experimental results verify the effectiveness of the proposed method in this dissertation.The dissertation has 76 figures,5 tables,and 167 references.