Micro-Doppler Analysis And Classification Of Human Activity Based On Continuous Wave Radar

With the development of society,in response to more and more emergencies and natural disasters,human body detection and activity recognition have gradually become powerful assistants for security,disaster relief,and monitoring.On the other hand,medical research shows that subtle changes in daily activity patterns and behaviors can help diagnose health problems.Compared with traditional video observation or instrument detection,radar has the advantages of all-weather operation,wide coverage,and good environmental adaptability.As an important basis for human detection and activity recognition in radar,the micro-Doppler effect has been extensively and in-depth studied in various application fields.This thesis has researched two aspects of human body micro-Doppler information extraction and analysis and human body activity classification.The main contents are as follows:(1)In this thesis,LFMCW radar is used to extract micro-Doppler information.Aiming at the phenomenon of time-varying micro-Doppler signals being submerged by noise,the MTI clutter suppression technology is simulated and selected,and the STFT joint time-frequency analysis method is selected through comparison to describe the micro-Doppler signal.The performance of the Doppler signal in the time-frequency domain.Theoretically,the existence form of micro-Doppler frequency shift in human body echo signal is deduced in detail,and the process of extracting micro-Doppler information by rearranging the sampling time of radar echo signal is explained.(2)Based on the Boulic human body simulation model,two human body models with large differences in height,body shape,and gait speed are designed.The micro-Doppler information is extracted and analyzed through simulation,which shows that the size and gait speed of the human body model are both compatible Affect the appearance of micro-Doppler information in the time-frequency domain.Based on the point scattering model,the micro-Doppler information caused by body parts such as the human body's torso,head,arms,legs,and feet are separately stripped out,and the periodic changes and performance of these body parts in the human body's micro-Doppler signal are analyzed.(3)The LFMCW radar platform was built,and multiple experiments were carried out.Concerning the analysis results of the simulation model,the human body micro-Doppler signal based on the measured data was analyzed in detail.The potential of the human body micro-Doppler detection system in extracting the micro-Doppler information of multi-person motion is explored,and a human body activity data set for activity classification is established.(4)Design two multi-classifiers based on the SVM algorithm for OvO and DT-SVMs models.Five activity feature parameters and three image features of HOG,LBP,and GLCM are selected.After classification experiments,the classification accuracy of the two models is more than 95%,the accuracy and compatibility of the OvO model are better than the DT-SVMs model,and the DT-SVMs model is better than the OvO model in terms of storage cost and training time.