Research On Human Posture Recognition Technology Based On Ultra-wideband Radar

Human motion recognition is widely used for smart homes,senior care,search and rescue operation,and surveillance.Due to the type of sensors used,the methods for human motion recognition can be generally divided into two kinds:wearable and non-wearable.However,there are many shortcomings for wearable sensors,such as poor portability,easily broken,low accuracy.The traditional non-contact sensors involve cameras,but cameras are always puzzled by lighting conditions and invasion of privacy.There are many advantages of human motion recognition based on radar sensor,such as no-contact,high accuracy and unaffected by lighting conditions,which has a great potential in human motion recognitioin for a long-term monitor.Since ultra-wide band radar has high resolution,strong penetration,and low power consumption,in this paper,human motion recognition based on UWB radar is achieved by the features extracting from the human motion echo signal.The main work is illustrated as follows:1.This article first introduces the human motion recognition system based on UWB radar,explains the principle of the transceiver module,the sampling module and the control module respectively,and makes a theoretical analysis of the human motion echo signal.2.Secondly,in the process of signal preprocessing,clutter suppression is applied to deal with the echo signals by using MTI and median filter.Because of various kinds of human motions in daily life,a pre-screening algorithm of human motion based on image processing is proposed to achieve a more accurate recognition.The speed and displacement characteristics of the human motions are obtained by means of image binarization,adaptive averaging filter and outlier removal algorithm,which is used to divide them into two general categories:Macro-motion and Micro-motion.3.Then,the weighted range-time-frequency transform(WRTFT)method is proposed.For the Micro-motion,features are extracted form the torso envelope in the Spectrum of WRTFT method.For the Macro-motion,principal component analysis(PCA)algorithm is used to extract the main components of the time-frequency representation(TFR).Next,the obtained features are sent to the machine learning module,Subspace K-Nearest Neighbor(KNN)model and the Bagged trees model are constructed respectively.Through the correspondence between the feature parameters and different human motions,human motions features are put into the classification model for recognition.4.Finally,the experimental platform of human motion recognition is established,and the experimental scheme is designed to validate the algorithm.Experimental data show that it can achieve an average accuracy of 94.4%for six kinds of Micro-motions by the Subspace kNN and an average accuracy of 95.3%for six kinds of Macro-motions by the Bagged trees,the human motion experiment results verify the effectiveness of the system.