Research On Wireless Intelligent Sensing Technology Based On Channel State Information

Based on the widespread rise of Wi-Fi sensing technology,Inspired by Wireless sensing theory,Thesis is derived from Channel State Information(CSI),the main research context is the size of sensing granularity,conduct in-depth research on coarse-grained position trajectory and activity behavior sensing,fine-grained gesture recognition,and Microfine-grained vital signs sensing,Aims to explore the sensing mechanism based on CSI signals at different granularities,establish the model relationship between the characteristic signal and the sensing target,and construct a Wireless sensing model.For different sensing environments,research methods and technologies such as CSI signal acquisition,data preprocessing,feature extraction,and sensing recognition.It brings new enlightenment to the Wireless sensing technology and algorithm of new wireless communication,and provides new technologies for the theoretical problems and application technical challenges faced by the application of Wireless sensing in the intelligent environment.Actively promoted the related applications of Wireless intelligent sensing technology in the fields of human-computer interaction,intelligent driving,health care,and action and behavior recognition and others.The main work is reflected in the following points:(1)Aiming at the problems of complex fingerprint generation and update,poor real-time positioning,high positioning error,clock out of sync and multipath effect in complex NLOS environment,an indoor positioning method based on CSI amplitude endpoint specific clipping and support vector machine(EC-SVM)is proposed,Firstly,the density based clustering algorithm is used to eliminate the outliers caused by multipath effect,through specific clipping(EC)of the CSI amplitude endpoints,and then fusion of the signals of the three CSI communication links,the feature extraction of the fusion links is performed to construct a fingerprint database.Finally,according to the cropped CSI position amplitude characteristics,SVM is used for classification,eventually the estimation result of the physical position is obtained.Experimental results show that the proposed method has good performance in real-time localization,trajectory tracking and intrusion detection of indoor moving targets.When the positioning error is 1.5 m,the positioning accuracy of EC-SVM algorithm can reach 89%.(2)Aiming at the problem that the accuracy of motion recognition is seriously affected by the movement direction of human body,it is difficult to guarantee the robustness of recognition in different directions,and it is difficult to recognize complex motion.A direction-independent motion recognition method(Wi-M)is proposed,Firstly,the CSI information of action behavior is collected by commercial Wi Fi devices,and the discrete wavelet transform is used for noise reduction,Then use principal component analysis and short-time Fourier transform to extract the Doppler frequency shift of the motion data,construct a frequency domain energy indicator,Taking Doppler frequency shift and fast Fourier transform value of human motion in frequency domain as common motion recognition features,Finally,the actions are classified and identified based on the long-short cyclic network.This method integrated spatial features into the time model,improved the robustness and accuracy of wireless signal recognition of human action,can effectively reduce the influence of the direction information of the action behavior,judge the start of the action,and has good environmental mobility and Recognition ability.Experiments are carried out in two kinds of common indoor environments(hall and office).The average recognition rate can reach 90.6% in different environments,and the average recognition rate can reach94.68% for head,hand,leg,trunk and other different parts.(3)Aiming at the universal problems of traditional gesture recognition,such as high cost,complex device operation,and strong intrusion,and how to solve the key issues such as the generalization ability of gesture recognition,the overall performance of gesture recognition,and the interactive recognition of multi-person different gestures.A CSI-based gesture language action recognition method is proposed,Collect the original gesture language data of the human body through the Wi-Fi device,use the Gaussian filter and the moving average filter to process the environmental noise in the original data,calculate the energy of each subcarrier,select the optimal subcarrier,and extract the gesture based on the time domain information of CSI Wave contour,select the mean,variance,skewness,and kurtosis of CSI gesture language action data to extract high-level features related to gesture language actions,Finally,the improved Adaboost classifier is used to classify and identify different gesture language actions.The performance of this method in handwritten digital gesture recognition is verified in real scene,and the recognition accuracy error is less than 2%;The concurrent sign language recognition in the scene of two to three person interactive conversation is realized,and the accuracy is more than 85%.It provides a feasible scheme for the application of Wireless sensing technology in gesture recognition.(4)Aiming at the difficulty of data acquisition and recognition of human movements with more subtle features like breathing,the basic principles of Fresnel zone perception model,including Fresnel zone reflection model and diffraction model,are studied,and the physical model of vital signs monitoring is established.A method of human breathing pattern detection based on CSI signal is proposed,This method uses the channel state information extracted from Wi-Fi signals as a measurement index to detect subtle breathing movements,and uses the difference in breathing and heartbeat frequencies to separate breathing and heartbeats.the Fresnel zone model is introduced to extract CSI respiration signals with obvious changes,and then the collected data is processed by Hampel filtering for abnormal values,and then The PCA algorithm is used to extract the optimal sub-carrier,and the sym8 wavelet function is used for further denoising and smoothing.Finally,The bidirectional recurrent neural network(Bi RNN)is used to construct a breathing pattern classifier to classify and perceive the signals of 4 different breathing patterns,so as to judge the current physiological state of the human body.The performance of the system is tested in two real scenes,and the overall recognition rate reaches 94.6%,which shows that the method has high recognition performance and strong robustness.