Wi-Fi Based Multi-target Gait Recognition And Single-target Diagnosis

Wi-Fi based gait matching has become a popular choice due to its noninvasive and ubiquitous.Existing intrusive multi-person gait matching approaches usually require live cameras and wearables,which generate privacy and feasibility concerns to subjects.However,without sensor,a major challenge comes with matching the reconstruction gait signals and orignals,i.e.,whether the gait and routines of subjects follow the standard baselines.In this paper,we address this limitation and propose a system,a Wi-Fi based multi-target gait recognition and single-target dagnosis system.The System can separate multi-user mixed gait signal with only one-pair transeivers.The contributions of this paper are as follows:· Multi-person mixed signal separation.When two persons walk along paths in pre-set path,the Wi-Fi signal reflected by the human body will be mixed in the CSI measurements at the receiver due to the CSI information collected will be in-fluenced by multiple subjects and produce many similar frequency components,which may be superimposed or subtracted.In order to seperate mixed-signal,based on both rigorous theoretical analysis and real-life experiments,we model multi-person gait matching as a BSS problem,which can be efficiently solved by the ICA method.· Segment motion signal.The signals we collect generally contain part of the interference phase without action or redundant action.How to extract the target signal we need from the continuous action signal will also have a great impact on our matching.In this article,we introduce an energy indicator to capture the significant change in CSI amplitude difference for automatic segmentation of signal segments with motion.· Spectrogram translation.Since the disentangled signals and the signals in the single-person environment are the same walking route of the same subject,they are still affected by the environment.In order to realize the transformation of features,we use Cycle GAN to translate the spectra of the disentangled signals into the spectra of the signals in the single-person environment.Then CNN is used to classify and identify.