Research On Moving Human Target Based On Millimeter-Wave Radar

The research of radar system to identify human targets has been widely carried out in many fields,including national defense,medical and sports applications.To recognize and classify a target,micro-Doppler signatures produced from various motions of a target can be a key feature for exploitation.In the early research stage,the classification features were mainly designed by applying the knowledge of the specialized domain,which limited the scalability of the algorithm.Therefore,we consider using deep learning methods to overcome this limitation.In this thesis,millimeter wave radar is used to detect the moving human body and collect the reflected echo.The time-frequency analysis is used to process the data and extract the micro-Doppler features.The deep learning algorithm is directly applied to the micro-Doppler spectrum to achieve the classification of human movement.The key work and accomplishment are as follows:1.The micro-Doppler effect and human motion model are studied.Through theoretical analysis and simulation,the human radar echo is obtained,and the simulation results are basically in accordance with the experimental results,which provides a theoretical basis for human movement detection and micro-Doppler feature extraction,and verifies the possibility of human motion classification using millimeter wave radar.2.The preprocessing method of human motion echo based on FMCW radar is studied.Six kinds of common human movements are designed in the experiment.The77 GHz millimeter wave radar is used to transmit and receive signals,and the micro-Doppler spectrum sample data of human movement is established.3.Aiming at the problems of complex feature extraction method and inadequate use of data in traditional classification ways,a deep learning method is proposed for classification and recognition of micro-Doppler spectrum,which is helpful for automatic feature extraction and improves the recognition rate of the algorithm.4.Aiming at the problem of large number of convolutional neural network parameters and large training samples,capsule network and deep capsule network based on residual structure are applied to human motion classification.The algorithms work well on the dataset of this thesis.The average recognition rate of capsule network is97.0%,which is 1.4% slightly higher than that of the convolutional neural network.The average recognition rate of deep capsule network based on residual structure is 98.5%,which is 1.5% a bit higher than that of capsule network.The above research work has been proofed by theoretical analysis,simulation test and experiment,which can realize the feature extraction and classification of human motion,and can be used to solve the problem of millimeter wave radar detecting human target movement characteristics.