Research On Contactless Human Activity Sensing Techniques Based On WiFi

Ubiquitous computing has strongly promoted the development of Io T technologies,which is the new direction of Io T sensing technology.Ubiquitous sensing technologies can be classified into contact sensing and contactless sensing according to their interaction with humans.Contact sensing technology is mainly deployed on the surface of the human body to obtain different parameters and analyze relevant parameters.Although contact sensing can obtain highly accurate sensing results,it requires the sensing target to carry the corresponding device at all times,which may cause inconvenience in daily life.In contrast,contactless sensing techniques do not require users to carry devices and can achieve continuous sensing,thus gaining wide attention from researchers.Among the contactless sensing efforts,WiFi signals have inspired a series of new sensing applications,such as trajectory tracking,activity recognition,identity authentication,and health detection,due to their strong sensing capability,wide communication range,and low device deployment cost.The previous WiFi-based human sensing work mainly used Received Signal Strength Indicator(RSSI)for target sensing,for example,by constructing RSSI fingerprints to locate people indoors.However,RSSI has the disadvantage of non-linear fading,which makes it impossible to perform high-precision sensing work.Compared with RSSI,the Channel State Information(CSI)of WiFi can characterize the channel characteristics of multipath signals in the surrounding environment and is sensitive to changes in the surrounding environment,enabling fine-grained sensing of targets.Although CSI-based human sensing work can perform high-precision perception tasks,existing methods still need to improve on problems related to single sensing scenarios,isolated sensing applications,insufficient reliability,and low accuracy.To address these problems,this thesis investigates a contactless human activity sensing method based on commercial WiFi and develops multiple sensing applications step-by-step.In summary,the main contributions of this thesis are as follows :(1)WiFi-based through-the-wall crowd counting method: typical WiFi sensing methods are usually worked for a single target,and the accuracy of the corresponding detection is severely decreased as the number of people increases.Therefore,crowd counting based on WiFi is the basis of other sensing works,especially in the through-the-wall scenario,which increases the difficulty of detection.To solve this problem,this thesis designs a WiFi-based through-the-wall crowd counting system,TWCC.This system corrects the CSI phase difference,and then extracts multidimensional signal features from four domains(namely,time domain,subcarrier domain,frequency domain,and time-frequency domain)to construct multivariate feature clusters.Combined with the multi-input neural network model,the TWCC system achieves accurate and reliable through-the-wall crowd counting.(2)WiFi-based indoor handwriting recognition method: The human-computer interaction problem is one of the classical works in wireless sensing.When the indoor person is a single target,the interaction task with the terminal is completed by sensing the person’s handwriting input in the air,It can significantly improve the efficiency of the interaction between the person and the intelligent terminal.To achieve this goal,this thesis designs a WiFi-based handwriting recognition system,Wi Reader.The system proposes a new CSI Ratio model by analyzing the interference caused by human arm motion to WiFi signals and combines CSI linear correction to extract the signal components influenced by the human body from wireless signals.Then the energy feature matrix is constructed using multilayer Discrete Wavelet Transform(DWT)and combined with Long Short-Term Memory(LSTM)network in deep learning to complete the recognition of handwritten English uppercase characters.We tested the Wi Reader system in several different scenarios,and the results show that Wi Reader can achieve highly accurate and robust handwriting recognition tasks.(3)WiFi-based human respiration waveform generation method: Human respiration is a more subtle and weaker physiological activity than large-scale activities such as handwriting.The chest rises and falls as a person breathes,causing a change in the propagation path of the wireless signal.However,it is easily masked by body movements and environmental noise.To sense human respiration,this thesis designs Breathe Band,a WiFi-based respiration waveform generation system.This system eliminates the interference of the background environment by using a multi-antenna CSI-subpopulation genetic algorithm,and extracts the human respiration components from the original CSI using independent component analysis.Finally,we propose a Mixing Cluster Gaussian-Hidden Markov Model to generate fine-grained human respiration waveforms.Extensive correlation experiments show that Breathe Band-generated respiration waveforms are highly similar to those captured by commercial respiration monitoring tapes under various conditions.It can achieve fine-grained and robust human respiratory monitoring.(4)WiFi-based human identification method: Breathing is a unique physiological characteristic of the human body,related to a person’s height,weight,bone,and lung structure,and can be used as a personal token.In this thesis,we design a WiFi-based human identification system,Breath ID.This system uses commercially available WiFi devices to detect human breathing and thus identify the user.The Breath ID system first performs vector analysis of the CSI ratio model and uses wavelet transform to obtain the breath features.Then,the human’s respiration rate is estimated by proposing a false peak elimination algorithm.Finally,a weighted multidimensional dynamic time warping algorithm is used to implement person identification based on the waveform of breathing features.Extensive real experiments validate that the Breath ID system can accurately identify multiple people and calculate the respiration rate of the sensing person.