| Under the background of the rapid development of communication and computer technology,especially the advent of the Internet of Things era,human beings have become the center of the information world.To realize the interaction between people and various devices,it is necessary to effectively perceive human behaviors.Current sensing technologies can be mainly divided into two categories:sensor-based sensing and radio-frequency-based sensing.This thesis focus on the WiFibased sensing technology in RF sensing.On the basis of an overview of its existing research work,it will focus on two implementations of human behavior sensing applications based on Channel State Information(CSI).In the process of human-computer interaction,personnel identification is the premise to ensure safe and reliable interaction.Traditional human identification usually requires dedicated sensing equipment,which limits the application scope of human identification.In order to realize non-devicedependent human identification,this thesis uses WiFi signal to collect and identify the breathing characteristics of people,and proposes a human identification system based on breathing perception.Firstly,theoretically analyze the CSI conjugate multiplication model and personnel breathing model;secondly,combine median filtering,Empirical Mode Decomposition(EMD)algorithm and Fast Fourier Transform(FFT)-based subcarrier selection strategy to separate the breathing component in the CSI signal;then,for the multi-antenna CSI data stream,the Constant False Alarm Rate(CFAR)peak-finding algorithm is used to detect the breathing rate;finally,the sliding window algorithm is used to generate the data set,human identification is done by using Convolutional Neural Networks(CNN).The experimental results show that the average absolute error of the system in detecting the breathing rate is about 0.5bpm,and the accuracy rate of human recognition in three scenarios has reached more than 85%,which meets the requirements of human recognition.In human-computer interaction technology,efficient sign language recognition can greatly improve the interaction ability.The current mainstream automatic sign language recognition input devices mainly include data glove-based sign language recognition and vision-based sign language recognition.Their common shortcomings are high equipment cost and inconvenient deployment,which makes it difficult to achieve large-scale popularization of deaf people.Based on the ubiquitous WiFi signal,this thesis designs a device-independent sign language recognition system.First,the CSI data is denoised by the Hampel filter;then the available subcarriers are screened by a selection strategy based on Dynamic Time Warping(DTW);finally,a deep learning model based on CNN and Bi-directional Long Short-Term Memory(BiLSTM)is used for sign language classification.The experimental results show that for the given 6 sign languages representing different meanings,the average sign language recognition accuracy of this system can reach more than 90%. |
