Research On Practical Technology Of Wi-Fi-Based Human Behavior Monitoring

In recent years,Wi-Fi sensing technology has become a research hotspot.It has important research significance and application value to introduce it into the field of behavior monitoring to realize the health care of the elderly.However,the existing research works on Wi-Fi sensing are mostly analyzed and verified from the perspective of theory and principle,and there are still many limitations in their practical applications.Focusing on the key research directions in the field of Wi-Fi human behavior monitoring,this paper designs the more effective mechanisms to meet the human behavior monitoring requirements in realistic scenarios from two aspects of human pose estimation and abnormal behavior detection,and builds the prototype experimental systems to verify the effectiveness of the proposed mechanisms.Aiming at the problems of device favorable deployment restriction and small sensing space in the scenario of human pose estimation,this paper proposes a link selection-based cooperative sensing mechanism.First,according to the effectiveness of Wi-Fi links for sensing human poses,all links are divided into Noise-Dominated Links,Most-Effective Sensing Links and Redundant Sensing Links.On this basis,a dynamic link selection algorithm is proposed.A threshold filtering method is used to remove the Noise-Dominated Links,and then an optimization problem is formulated to select the Most-Effective Sensing Links from the rest Wi-Fi links.Finally,the CSI data corresponding to the selected links is input into the neural network to estimate the 3D coordinates of the key points of human pose.The experimental results show that the average joint position error of the system on the 4 subjects participating in the training process reaches 32.31 mm.It achieves high-precision human pose estimation everywhere indoors,reduces the computational redundancy,and has a good cross-target generalization ability.Aiming at the problems of unpredictability,rare occurrence and difficult data collection of human abnormal behavior in the scenario of human abnormal behavior detection,this paper proposes an unsupervised learning-based feature reconstruction mechanism.This mechanism builds a network structure consisting of a generative network,a reconstruction network and a discriminative network,and designs a joint loss function.In the training process,a large number of CSI data samples of normal human behavior are input.By first reconstructing the original input,and then reconstructing the feature vector of the latent space,the neural network has the ability to accurately reconstruct the characteristics of the human normal behaviors.In the detection process,by measuring the difference before and after the reconstruction of human behavior characteristics,combined with the comparison of abnormal score thresholds,it is determined whether the abnormal behavior has occurred.The experimental results show that the system can effectively detect non-specific human abnormal behaviors,and achieves a detection rate of 96%and a false alarm rate of 13%,which proves that the mechanism achieves excellent and practical performance on human abnormal behavior detection.