Indoor Human Sensing Based On Wireless Signals

Smart sensing technologies can obtain key information including human status and location,which is an important bridge between human and information network.The non-contact and all-weather nature of wireless based systems makes it have the potential to become the fundamental building block of next generation sensing systems.However,the complexity of wireless signal itself has brought great challenges for accurate wireless sensing.In this dissertation,we conduct theory and application research focusing on the key technology of wireless sensing.The main contributions of this dissertation are summarized as follows1.We propose calibration methods to remove phase offsets in WiFi signal,which makes WiFi signal be applicable to sensing applications.To remove different kinds of phase offsets,we combine theoretical analysis and experiment verification,and propose calibration methods based on angle estimation,reference signal and paths-conjugation,respectively.2.We propose a WiFi- based human breath tracking system.To resolve the challenge introduced by the limited spatial resolution and severe noise and interference of WiFi devices,we propose a weak signal extraction method based on sparse re-construction,which could effectively aggregate the information on WiFi devices,suppress multi-path interference and achieve fine-grained breath status tracking.3.We propose a human location and vital signs tracking system based on a wideband multi-antenna system.To resolve the complex propagation problems in indoor environ-ments,we analyze the features of different multi-path interferences,and propose a human tracking method based on shortest-path algorithm and a multi-path recognition method based on unsupervised clustering.Based on these methods,we implement a location and vital signs tracking system.4.We propose a millimeter-wave human location tracking system based on cross-modal supervision.To resolve the strong interference in practical deployment for millimeter-wave systems,we propose to collect wireless and video data simultaneously and adopt the processed video data as the supervision for wireless data.Based on this method,we build a dataset,train a deep neural network and finally achieve accurate human localization.5.We propose a liveness detection method by fusing the ultrasonic and visual in-formation.To resolve the challenge that the existing video based systems are vulnerable under attack,we introduce the ultrasonic modality.During liveness detection,we let user perform lip movement,i.e.,open and close mouth.Based on the modulation on ultra-sound reflection caused by lip movement and the video data,we achieve accurate liveness detection by fusing the information of two modalities and enhance the security of facial authentication systems.