Research On Human Motion Recognition Method Based On Millimeter Wave Radar

In recent years,with the rapid development of artificial intelligence,human motion recognition has become the focus of much research.Although wearable-based human movement recognition methods have a high accuracy rate,they require the wearer to wear a series of kinematic sensor devices,which is not only heavy and inconvenient to wear,but also costly.Vision-based recognition techniques have good accuracy and are less costly,but are susceptible to interference and significantly reduced performance when applied to harsh environment scenarios.Millimeter wave radar-based human motion recognition has become a popular research direction in academia and industry due to its high interference immunity,high resolution,and excellent protection of personal privacy.This paper presents a study of human motion recognition based on Frequency Modulated Continuous Wave(FMCW)radar,as follows.Firstly,the process of human action signal processing based on millimeter wave radar is introduced,the IF signal of millimeter wave radar is analyzed and derived,the distance and speed measurement method is described based on the IF signal,and the parametric resolution of the radar is explained.The FMCW millimeter wave radar is used to process and analyze the echoes of human gestures and human body movements,and to generate a Range Doppler Map(RDM)based on the relationship between distance,velocity,and frequency of the IF signal.Secondly,this paper investigates a deep learning-based method for human gesture distance-Doppler heat map feature extraction and proposes a new IC3 D network(Improved 3D Convolutional Neural Network,IC3D)for dynamic gesture heat map feature extraction.While reducing the number of parameters in the convolutional and fully connected layers,the SGD algorithm in the original network is replaced by the Adam algorithm,and finally,the extracted gesture features are stitched and classified in the softmax layer.The experimental results show that the average recognition accuracy of the IC3 D network proposed in this paper is 99.8% for each gesture,and the computational delay is less than 80 ms.Finally,a methodological study of human limb movement recognition is carried out.Firstly,the system components of human action recognition are introduced,and then a distance Doppler time series detection network(3DTSNet)model of a three-dimensional convolutional neural network is proposed.A smooth non-monotonic function,Mish,is used as the activation function of the 3DTSNet model.Finally,different human limb movement radar datasets are used as input samples for training the 3DTSNet network.The experimental results show that the detection accuracy of the proposed 3DTSNet network model is over 98% and has good generalization.