Research On Continuous Motion Recognition Technology Based On Linear Frequency Modulation Continuous Wave Radar

With the development of artificial intelligence,human motion recognition has gradually attracted the attention of scholars at home and abroad.The main task of human motion classification is to extract human motion information from radar echo signals and to classify human motion by signal processing and machine learning.This technology has broad application prospects in smart home,safety monitoring,elderly guardianship and other industries.The existing motion recognition algorithms mainly focus on the recognition of single motion,but in practice human motions are always continuous.Therefore it is very important to recognize continuous motions.This paper utilizes FMCW radar to collect human motion echo signals.The characteristics of range-Doppler domain are extracted from six kinds of motions including falling,stepping,jumping,squatting,running and continuous motions after free combinations.The machine learning classifier is used to realize motion classification.Details of the work are given as follows:1、Principles of radar range and velocity measurements are introduced.The composition of the continuous motion identification system and the principle of each module of the system are described,and the process of radar echo signal preprocessing is presented.2 、 Traditional motion identification algorithms are introduced,including principal component analysis based motion identification algorithms,time-frequency map based motion identification algorithms and range-Doppler map based motion identification algorithms.Principles of these algorithms are introduced in detail,and the performance of each algorithm is analyzed by using measured data.3、A method of continuous frame range-Doppler transform is proposed.The signal is divided into several frames in time domain.The range-Doppler transform is applied to each frame to obtain the dynamic range-Doppler image of human motion signal.In each frame of range-Doppler image,the maximum energy point is selected to extract the dynamic rangeDoppler trajectory of human motion.The appropriate energy threshold is selected to determine the starting frame of the motion.The coordinates of each point in the dynamic range-Doppler trajectory and the dispersion of each frame are extracted by sliding window method as features.The motion is identified by using the model of machine learning classification algorithm.4、The experimental scheme is designed and the data of eight experimental subjects are collected.For single motions,the experimental data of different distances,angles and directions are collected to verify the robustness of the algorithm.For continuous motions,the data of two consecutive motions and the data of free movement with multiple motions are collected respectively.The experimental results show that the proposed algorithm is effective.The proposed continuous motion recognition algorithm uses Subspace KNN for continuous motion identification with the final accuracy of 93.3%,which verifies the effectiveness of the algorithm.