Line-of-Sight Human Flow Detection Base On Channel State Information

With the rapid development of Internet of Things technology and the widespread deployment of WiFi networks,many infrastructure devices have brought a lot of information resources to people.Using these information resources to further improve people's lifestyles has gradually become the future trend of development,enabling the environment sensing technology which not only uses the WiFi signal to sense the surrounding static environment,but also realizes the corresponding customized services by sensing the environment change,such as human detection,gesture recognition and activity recognition.Traditional environment sensing technologies can be categoried as sensorbased,RFID-based,RSSI-based and CSI-based technologies,among which CSI has been proved to provide better environment perception ability than other signals by researchers since it can distinguish different multipath components,making it a finer-grained descriptor of the wireless channel.In this thesis,existing environmental sensing technology is firstly studied in depth,then several common environment sensing techniques are summarized,with their advantages and disadvantages described in detail.Afterwards,CSI-based line-of-sight path human number detection scheme is designed,including line-of-sight path identification and queue number detection.In the CSI-based line-of-sight path identification method,the experimental scene is divided into two scenes,i.e.,static scenario and dynamic scenario respectively.In the static scenario,a group of feature cluster is extracted and then classified by using back-propagation neural network.In the dynamic scene,the K-Mean feature is extracted with the Rician-K model to realize dynamic line-ofsight identification.In the above two scenarios,the average recognition rate is around 95%.In the CSI-based queue number detection method,this thesis first uses moving average method to denoise raw CSI signals.Then,two features of skewness and kurtosis are extracted,and the twodimensional feature is input into a radial basis-based support vector machine to realize the identification of queue number.Finally,through the existing concept of Fresnel Zone,the number of queue is dynamically adjusted.The experiment results show that the accuracy of the number detection can be maintained at about 90% with high robustness in different scenarios and conditions.