Human Identification And Behavior Sensing Based On Millimeter Wave Radar

Developing remote health monitoring system is one of solutions of alleviating the pressure of an aging population,whose three key elements are body positioning,activity status estimation and human identification.In existing behavior monitoring technologies,radar has the advantage of non-contact,non-invasive,light insensitivity,long distance,high accuracy and resolution.This paper introduces a system that is able to achieve body positioning,activity status estimation,and human identification based on FMCW(Frequency Modulated Continuous Wave)radar.For the subsystem of body positioning and activity status estimation,the challenges come from interference of noise in the environment caused by static objects and dynamic objects such as air-conditioning and fans.We obtain position and velocity of objects by transforming raw data into Range-Doppler spectrum and Range-Angle spectrum by Fourier transformation,static cluster removal algorithm and object detection.The key idea of activity status estimation is that position information contains activity status and non-human objects are removed based on their trajectories and patterns of Doppler.The accuracy of activity status estimation is 82.7%.For subsystem of human identification,we obtain Micro-Doppler spectrum by short-time Fourier transform,which represents gait character.Further,the gait features are analyzed,and then SVM(Support Vector Machine)and CNN(Convolutional Neural Network)are used for classification.Otherwise,open set is introduced because the system should reject unknown people.We design experiments and test the result of solving open set problem using Siamese network.The accuracy of classification using Siamese network is 98.47%,which is better than using SVM and CNN,and the AUC(Area Under Curve)of human authentication using contrastive loss for training is greater than 0.93.