Research On Wi-Fi-based Contactless Monitoring For The Elderly

As the global population ages,the elderly are prone to physical decline and sudden illnesses,making them vulnerable to emergencies in their daily lives.Symbiotic sensing and wireless communications provides a new solution for the elderly monitoring.Wi-Fi signals are ubiquitous in indoor environments and carry information reflecting the features of the propagation space.Using Wi-Fi signals to monitor the elderly,detect emergencies timely,determine the triggering causes accurately can save time for treatment and minimize risks.In this dissertation,Emergency Semantic Feature Vector(ESFV)composed of Position,Behavior and Respiration features are extracted from the ubiquitous indoor Wi-Fi signals to achieve contactless,cost-effective,non-invasive monitoring and accurate,effective emergency notification for the elderly.This dissertation conducts an in-depth study on how to obtain effective sensing data with ubiquitous Wi-Fi devices and how to accurately extract the Position,Behavior and Respiration features of ESFV.The main research results and innovations of this dissertation are summarized as follows.1.To obtain effective sensing data,this dissertation proposes an efficient collaboration strategy of ubiquitous Wi-Fi devices.Based on the effectiveness of Wi-Fi links to human activities in indoor environments,this dissertation divides Wi-Fi links into interference links,most-effective links and redundant links,and proposes the Dynamic Link Selection(DLS)strategy to remove interference links and select most-effective links.This strategy maximizes the link importance and minimizes the redundancy.In this dissertation,most-effective links are selected for 3D human pose estimation,and experimental results demonstrate that the DLS strategy can efficiently and accurately locate 3D human keypoint coordinates with multiple Wi-Fi links exist.The efficient collaboration strategy of ubiquitous Wi-Fi devices provides effective sensing data for ESFV extraction.2.To accurately extract the Behavior features,this dissertation proposes a 2D human pose estimation method and a 3D human pose estimation method based on Wi-Fi.The 2D human pose estimation method is based on the proposed resolution improvement and human-related signal extraction methods.A neural network is designed to map the Wi-Fi signals into human pose skeleton images,and a generalization ability improvement mechanism is proposed to enhance the generalization ability of the 2D human pose estimation model to new human subjects.A prototype system is built,experimental results verify that the method can achieve similar human pose estimation performance as that of cameras,and can even estimate human poses through a wall.The 3D human pose estimation method is also based on the resolution improvement and human-related signal extraction methods.A neural network is designed to map Wi-Fi signals to 3D human keypoint coordinates.A prototype system is built,experimental results show that the method can locate the 3D human keypoint coordinates with better performance than other methods in both Line of Sight(LoS)and None-Line of Sight(NLoS)cases.ESFV Behavior features are accurately extracted from the human pose estimation results.3.To accurately extract the Position features,this dissertation proposes a Wi-Fi-based human tracking method.The method uses the ratio of Channel State Information(CSI)between two adjacent antennas to denoise the original CSI,and further separates the signals and noises over subcarriers using Principal Component Analysis(PCA).To reduce the influence of static components,a robust signal calibration algorithm is designed in this dissertation to improve the tracking accuracy.The trajectory tracking is achieved by analyzing the relationship between the phase of the CSI ratio and the dynamic path length.In this dissertation,a prototype system is built,experimental results verify that the proposed scheme can achieve indoor human tracking at the decimeter level.ESFV Position features are accurately extracted from the human tracking results.4.To accurately extract the Respiration features,this dissertation proposes a Wi-Fi-based human respiration status monitoring method.Based on the mapping model of CSI amplitude/phase stream and human respiration status,this dissertation defines Respiration-to-Noise Ratio(RNR)to select the amplitude/phase stream with the most sensitive sensing ability to obtain the respiration status curves.In addition,the method detects abnormal respiration status(apnea)and classifies abnormal respiration patterns(e.g.,ataxic respiration and cheyne-stokes respiration,etc.)based on the respiration status curves.This dissertation builds a prototype system,experiment results verify that the correlation coefficient between the respiration status curve obtained by the method and the ground-truth reaches 80%,and the accuracy of apnea detection reaches 91.2%.ESFV Respiration features are accurately extracted from the human respiration status monitoring results.