Research On Human Posture Recognition Method Based On FMCW Radar Signals

In recent years,relying on people’s living environment tends to be indoor,people’s demand for indoor intelligent home,intelligent office,intelligent entertainment and so on is also increasing,which makes human posture recognition technology has application prospects,there have been a variety of human posture recognition technology has been applied to different scenes.At present,human posture recognition technology has been studied in vision,wearable sensor,visible light,acoustic wave,Wi Fi and other aspects.However,the existing pose recognition methods have some limitations.First,the human posture recognition method based on vision,wearable sensors and visible light is easily blocked by obstacles and affected by temperature and illumination.Second,the body posture recognition method based on sound wave and Wi Fi will not be affected by illumination,but it has high power consumption and is easy to be interfered by external environment.In order to solve the above issue,this essay put forward a research way of human posture recognition based on FMCW radar signal,which effectively solves the influence of adverse natural prerequisite such as strong light,darkness,temperature and so on.The concrete research contents are as follows:(1)The research of FMCW human posture recognition method based on background noise elimination is put forward.This paper mainly uses Texas Instruments(TI)company’s FMCW millimeter wave radar module to identify human posture.Firstly,the discrete Fourier transform(FFT)is carried out on the data collected by FMCW radar using MATLAB software,so as to obtain the target distance,target velocity and target Angle.Secondly,DBSCAN clustering algorithm and Hampel filtering method are used to solve the noise interference of dynamic or static targets in the range,while removing redundant outliers to improve the accuracy of human posture,so as to construct Distance-Time(DTM)and VTM.Finally,in the python software,a multi-dimensional parameter deep learning network framework based on different fusion methods is built.The network framework uses the convolutional neural network method to extract the features of the DTM and VTM data sets,uses the feature series fusion method and the high-efficiency low-rank multimodal fusion(referred to as LMF)method to fuse,and uses the activity recognizer to obtain the classification results and the application domain.Discriminator.Therefore,this method effectively removes the background noise interference of FMCW human posture recognition,and is not sensitive to the environment.(2)The real-time realization method of human posture recognition based on FMCW radar signal is proposed.This article mainly relies on FMCW radar to collect different data,and uses computers and Jetson Nano devices to build an edge computing platform.The edge computing platform mainly includes the PC end and the edge end.The PC terminal transmits the human body posture data collected in real time by the FMCW radar to the edge terminal through the local area network.At the edge,the deep learning network of the trained LMF is used to realize the real-time realization of the human body posture category.Experiments show that the posture recognition method proposed in this paper can recognize posture in real time.Therefore,the related research and theoretical results of this paper can be used to build and improve the accuracy of the human posture recognition system,realize the human posture recognition of FMCW radar in various states,and meet the needs of society.