Research On Passive Human Detection Techniques Based On Channel State Information

With the widespread deployment of wireless infrastructure,wireless sensing has attracted increasing attention from researchers.Compared with traditional camera and infrared based solutions,wireless sensing is independent of illumination change and has the advantages of wider coverage and privacy protection.Since it does not require users to carry any device or actively participate in the sensing process,passive wireless sensing has become one of the most popular techniques and fostered myriad emerging applications such as human intrusion detection,health monitoring and indoor localization.Serving as the primary task of passive sensing,passive human detection leverages wireless signal to detect human presence,which determines whether further sensing is performed or not,e.g.,behavior recognition or location estimation.Despite of great progress achieved,existing human detection methods usually only apply to specific scenario and suffer from limited detection range,accuracy and reliability.Targeting at the above problems,this dissertation studies channel state information(CSI)based passive human detection techniques and uses commodity off-the-shelf(COTS)Wi-Fi devices to design high-accuracy human detection schemes with low deployment cost and high robustness for different detection scenarios,which are all prototyped in our daily living environment to verify their effectiveness,enabling large-scale deployment and long-time operation of passive human detection system in real environment.Besides,based on the different human detection schemes proposed above,this dissertation further designs an adaptive unified passive human detection framework,which can choose appropriate detection scheme based on the detection scenario and target state adaptively,providing a feasible implementation for pervasive human detection system in real environments.In summary,the main contributions of this dissertation are as follows:(1)CSI-based passive indoor moving human detection.To deal with the unsatisfactory detection accuracy and range of existing works,this dissertation proposes a robust indoor moving human detection system R-PMD,which extracts the variance distribution of CSI amplitude as the environmental feature profile.Specifically,R-PMD decides environment status by computing the EMD distance between online feature profile and offline stored feature profile of human-free scenario.Results from real experiments verify the accuracy and effectiveness of R-PMD.(2)CSI-based passive through-the-wall moving human detection.To cope with the high deployment cost problem caused by dedicated devices and signals adopted in existing methods,this dissertation presents a robust through-the-wall moving human detection system R-TTWD with high detection accuracy,which is implemented on cheap COTS Wi-Fi devices.Unlike previous time-domain feature based solutions,R-TTWD leverages the correlated amplitude variations among different subcarriers in frequency domain caused by moving target and extracts the mean of the first-order difference of eigenvectors of CSI correlation matrix as the detection feature.Afterwards,R-TTWD trains the support vector machine classifier with only a few offline extracted features and employs the derived classification hyperplane to identify moving human presence or absence online.Extensive experiments in different real environments demonstrate that R-TTWD can not only gain high detection accuracy and robustness over time,but also be promising for realizing environment-independent detection.(3)CSI-based passive indoor static human detection.To further expand the practicability of existing detection systems which rely on the mobility of human,this dissertation introduces a robust static human detection system R-PSHD,which leverages respiration rate estimation to infer static human presence.Specifically,R-PSHD utilizes CSI phase information for respiration rate estimation and computes phase difference of different antennas to remove the intrinsic phase randomness of COTS devices.By adopting a series of signal processing techniques including subcarrier selection,R-PSHD can estimate the respiration rate accurately and thus perform static human detection according to the estimation result.Extensive real experiments validate that R-PSHD can improve both detection distance and system reliability.