Research On Device-Free Human Detection Based On Channel State Information

With the development of communication technology and the comprehensive coverage of Wi-Fi in cities,environmental sensing using Wi-Fi signals has gradually become a research hotspot.Gesture recognition and fall/breath detection have been widely applied in humancomputer interaction,property security,health monitoring and other fields.Wi-Fi-based method is inexpensive and can be widely deployed,and do not require the person being monitored to carry any special device.Compared with the method based on computer vision,the method has no requirement on illumination condition,no problem of visual blind area and privacy leakage,which provides great convenience for practical application.The main research contents are as follows:In this paper,a unified framework is proposed to realize the detection of dynamic targets and static targets at the same time.For the detection of dynamic targets,an algorithm based on CSI sliding variance is adopted to calculate the time period of human activities within the sensing range.For static targets,the Fresnel sensing model was used to obtain that CSI signals would show sine-like waveform along with the periodic movement of the chest cavity.Then,the robustness index was designed in combination with the degree of curve fitting and the estimated frequency to evaluate whether there is static target within the sensing range of the link.In order to improve the accuracy of detection,this paper considers the amplitude and phase difference of feature extraction,and designs a set of filtering,including outlier removal,linear interpolation,wavelet denoising,etc.The method of selecting data stream and subcarriers is also proposed,which can greatly reduce the computation while retaining most of the characteristics,and can well adapt to the situation that the channel state information occasionally appears unstable.In order to make the actual deployment of the system more convenient,this paper also proposes an environmental adaptive module to adapt to the complex and changeable sensing area and the deployment of multi-link and multi-scene.It only needs a small amount of measurement to determine the model parameters without the calibration of specific scene.