Recognition Of Road Moving Target Based On The Time-Frequency Characteristics Of Millimeter Wave Radar

Because of the low environmental requirements and high resolution,the millimeter wave radar is widely used in military fields and civilian areas.Accurate identification of road moving targets plays an important role in traffic monitoring,traffic command and smart traffic.In this thesis,the recognition performances of time-frequency image and radar cross-section(RCS)are studied,and the identification task of road moving targets is completed.The following work is carried out on the recognition of the road moving target:(1)A simulation module for obtaining target motion information is established.The motion information of the target is an indispensable condition to construct the target echo signal.Accurate motion information can make the echo signal more conformity with practical situation.In this thesis,two methods for obtaining target motion information are introduced,which are mathematical model and motion capture file.The mathematical model and motion capture files are used to obtain the motion information of the limbs of human and dog respectively.The shadowing effects of human limbs are analyzed and a more rigorous human motion model is established.(2)A radar simulation module is established.Radar simulation module constructs target echo signal by using target motion information and radar parameters.First,RCS is introduced in this thesis.What's more,the echo model based on linear frequency modulated continuous wave is analyzed,and the rang-velocity coupling is explained.Then the problem of rang-velocity coupling is solved by using the range-Doppler processing method.After range-Doppler processing method,the stationary targets are removed and the velocity resolution is improved.Finally,the performance of the range-Doppler processing method is verified by simulation(3)The advantages and disadvantages of several extraction methods of time-frequency image are analyzed.Time-frequency image is one of the important features of target recognition,which can be extracted from radar echo signal.The principles of time-frequency analysis methods,including short-time Fourier transform,pseudo Wigner-Ville distribution,S method and pseudo Wigner-Ville distribution,are first introduced in this thesis.The advantages and disadvantages of these methods are analyzed and simulated.Then,the time-frequency images of human motion are simulated by these methods respectively.Several important micro-Doppler features are also introduced in this thesis.In addition,the consistency of the time-frequency image is satisfied by correcting the Doppler shift of the main scattering point and the characteristics of the time-frequency image are more markedly after normalization.Finally,the hardware platform of millimeter wave radar used in this thesis is introduced.The processing of the measured data is described in detail,and the time-frequency images of the measured data are obtained.(4)Target recognition and classification are completed by using deep learning method.First,the basic unit and basic structure of neural network are introduced,and the principles of the convolutional neural network(CNN)are summarized.Then,the performance of the recognition method based on micro-Doppler and CNN is verified by the measured data.Next,the principles of the support vector machine(SVM)are summarized and the performance of the recognition method based on RCS and SVM is verified.Finally,a method based on the fusion of micro-Doppler and RCS is proposed.The excellent performance of fusion method is verified by the measured data.