Research On Human Feature Recognition Technology Of Broadband Multi-channel Millimeter Wave Radar

The lip recognition,gesture recognition,and human posture recognition of human body feature recognition have important application value in human-computer interaction.And there are very broad application prospects in the fields of smart home,somatosensory games,assisting the deaf and dumb,and security monitoring.At present,the more mature human feature recognition technologies are based on vision and contact sensors.However,visual technology is sensitive to light and cannot protect personal privacy.And contact sensors need to be worn additionally,which affects user experience.Yet the recognition of human body characteristics based on radar is not only immune to external environments such as light,weather,temperature,and sound,but also protects personal privacy.At the same time,the multi-channel millimeter-wave radar also has the ability to detect azimuth and pitch angles,which can completely acquire the three-dimensional characteristics of human movements.Therefore,this paper proposes a human body feature recognition technology using broadband multi-channel millimeter wave radar,moreover,builds lip recognition,gesture recognition and human posture recognition systems respectively.The main tasks completed in this paper are:1.Aiming at the problem that the optical image lip language recognition method is sensitive to light conditions and cannot protect personal privacy.This paper proposes for the first time a lip language recognition method based on radar micro-Doppler features,using radar micro-Doppler images with distinct lip movements as CNNs model recognition information to recognize different lip words.At the same time,in order to better integrate lip language features,the CNNs module introduces an AM mechanism,which can make the network model focus on more important lip language action signal areas and suppress background noise areas.This method makes full use of the micro movements of lip language including lips,tongue,and throat,and can fully reflect the characteristics of lip language movement.Through the comparison of experiments on different individuals and different working ranges,the experimental results prove the effectiveness of the method,and the recognition rate of 10 English letters exceeds 87.5%.2.Aiming at the problem of insufficient micro-Doppler feature angle information in the single-channel radar gesture recognition method.This paper proposes a micro-Doppler feature fusion of azimuth and pitch angle information as a gesture recognition method for CNNs model recognition information,and uses the multi-channel detection capability of AWR1642-ODS radar to obtain three-dimensional micro-Doppler features of gesture actions.In order to better integrate gesture features and suppress noise and clutter,gesture recognition also introduces an AM mechanism to improve the CNNs model.In addition,the method is also based on the gesture movement mechanism for data expansion.Extensive experimental evaluation proves that the multi-channel feature fusion method can significantly improve the recognition accuracy.For the established 10 gesture data sets,this method achieves a recognition rate of 95.7%,which is about 5% higher than the recognition rate of the single-channel method.3.Aiming at the problem that the light sensitivity cannot protect personal privacy and the lack of information about the two-dimensional posture feature of the human body in the optical image human posture recognition method.This paper proposes a human body attitude recognition method based on millimeter wave radar 3D point cloud,which uses the three-dimensional detection capability of IWR6843-ODS radar to obtain the point cloud distribution information of the human body in three-dimensional space.And based on the improved tangent midpoint algorithm,the 3D point cloud skeleton height feature and the top view boundary size feature are extracted respectively.Finally,the experiment established a data set for training and recognition by setting eight different human postures and simultaneously recording human body data of different heights.Experimental results show that the comprehensive recognition rate of the eight gesture changes reaches 95.6%.