| Life detection radar has been widely used in military anti-terrorism,disaster rescue,medical detection and other fields,and more and more attention has been paid to the extraction and identification of human body nutation.This paper is mainly based on two methods to detect human life under static and moving conditions.Life detection of a stationary human body based on microdoppler characteristics of heartbeat and respiration;In the case of human body movement,the micro-doppler characteristics of gait are mainly used to realize life detection,and the key problems are studied.Aiming at the modeling of motion model and radar echo model,this paper first establishes the sinusoidal model and the sharp pulse model of respiratory and heartbeat,and compares the advantages and disadvantages of the two models.On the basis of humanoid robot and virtual human research,using the biomechanical experimental data measured by m.t.halaman and d.t.halman,an empirical human motion model suitable for the study of human radar characteristics was established,and the kinematic trajectory of each limb was analyzed.Similar to human motion modeling,this paper used the CMU graphics laboratory database to obtain the kinematics parameters of horses,established a four-legged model coordinate system,and established a set of animal gait model by using the relationship between the flexural angles of each joint and euler rotation matrix.On this basis,the analytical expressions of radar echoes of various motion models are deduced,which provides a theoretical basis for human detection and feature extraction.Aiming at the problem of the extraction of respiratory heartbeat frequency,this paper adopts the breathing heartbeat model established in the second chapter,and uses a breathing and heartbeat frequency information extraction algorithm,namely two time frequency analysis and curve fitting,and the MATLAB simulation proves that the algorithm is feasible.Aiming at the problem of micro Doppler feature extraction,this paper explores a new time frequency analysis method,namely,synchronous extrusion short time Fourier transform(SSTFT),and finds that the algorithm has achieved good results in the field of vibration engineering.This paper applies it to the gait micro motion feature extraction for the first time,and through simulation,we can see that the time-frequency map after SSTFT transformation is higher than that of STFT transform.The rate of focusing is significantly improved,and the effect of micro Doppler feature extraction is better.Aiming at the problem of human motion recognition,a gait based feature extraction algorithm is proposed in this paper,which is to extract the average speed of the trunk,the frequency of the leg swing and the maximum and minimum speed of the limb swinging as the features of the classification,and classify them by machine learning algorithm.Finally,experiments verify that the recognition accuracy of human motion state(walking and running)based on the machine learning model trained by the feature extraction algorithm is as high as 95%,which is 6.25% higher than that of the traditional image feature(HOG feature)extraction algorithm.In order to support the above research work of this paper,this paper USES Infineon(Infineon)24GHz continuous wave radar to collect a large amount of experimental data,and extracts the micro-doppler characteristics of the measured human gait.By comparing the measured and simulated images,the motion model and algorithm theory established in this paper are experimentally verified. |
