Research On Human Target Detection And Feature Recognition Of LFMCW Radar

With low Radar Cross Section,slow moving velocity,it is a tough challenge for radar to detect and classify slowly moving human target whose Doppler frequency is overlapped by that of the strong ground clutter.In this thesis,an investigation of human detection and feature recognition using linear frequency modulation continuous wave(LFMCW)radar is given as follows:First,a simplified human body model with twelve joints and ten scatter points is designed.The rotation functions of twelve joints and displacement functions of ten points on the human coordinate for the simplified model are derived.Based on above derivation,echoes of LFMCW radar for human motion target are constructed.The correctness of the simplified human body model and the constructed LFMCW human echoes are verified by the comparison between the spectrogram of the simulated human echoes and the sampled data from an LFMCW radar system developed in our laboratory.Second,to solve the problem of Doppler spread during long time coherent integration for human detection using LFMCW radar,a method called phase compensation transform(PCT)is proposed.The PCT method matches the nonlinear phase of human beat signal by constructing phase compensation signal through iteratively searching optimal motion parameters,thereby Doppler spread energy is accumulated and the detection SNR is increased.The effectiveness of the proposed PCT method is validated by simulation and sampled data from LFMCW radar.Third,with respect to human motion pattern recognition using LFMCW radar,micro-Doppler signatures of simulated human target echoes and sampled data from LFMCW radar with different human motion patterns are studied by time-frequency analysis.Then human motion features are summarized from the above time-frequency results.A human motion pattern recognition method employing a support vector machine decision-tree structure is designed.Based on the sampled data from LFMCW radar,the human motion pattern recognition performance is evaluated.