Human Micro Behavior Sensing Based On Passive Wireless Backscattering

With rapid development of the Internet of Things(IoT),various IoT applications and technologies emerge continuously.Among them,two aspects of research have attracted attention from both business and academia.On the one hand,passive backscatter communication technology greatly reduces the transmission energy consumption,and thus breaks the energy bottleneck restricting the development of the IoT.On the other hand,IoT based on human behavior sensing effectively expands the application scenarios of the IoT,and serves as an important role for everything interconnection.In this paper,we organically combine these two aspects to realize human behavior sensing and user identification based on passive backscattered signals.Our work does not need additional equipment,and only takes the passive backscattered signal as the source of behavior perception.Therefore,it provides human perception ability for low-performance IoT devices along with the premise of low energy consumption,so as to effectively expand the application scenarios of IoT.In this paper,we focus on contactless human behavior sensing based on passive backscattered signals,with the target for finer and more effective sensing technologies from two dimensions,i.e.,action granularity and semantic complexity.We propose a series of perceptual recognition and identification for human micro activities.Detail of our work is shown as follows:For the scenarios with small activities and fixed semantic,we propose Word-Fi,a handwriting recognition system.We design an adaptive segmentation algorithm to accurately slice the input signal.Through feature extraction in both time,frequency and phase domains,the recognition rate of 26 capital English letters handwritten by people can reach 90%.Further with error correction and word formation of the recognized string,we can achieve word recognition rate of 96%.In the multi-person scene,our passive backscattered signal exhibits its advantage of small signal radiation range.After filtering and removing interference,it can support multi-person interaction at the same time in a short distance.For the scenarios with micro activities and fixed semantic,we propose a human fingertip activities recognition and user identification system denoted as Finger-Fi.We further reduce the perceived action scale,and realize the perceptual recognition of fingertip level actions.We also implement user identification by analyzing the characteristics of fingertip actions of different people.We set up two common application scenarios according to the practical usage,and propose high-precision action segmentation algorithms adapted to different application scenarios.Through the feature extraction of signal in time domain,frequency domain and change rate,we realize the recognition of 26 lower case English letters,and the accuracy is about 93%.Meanwhile,we also achieve user identification with accuracy of 95%by analyzing the activity characteristics of different people.Our work provides a new interaction method for elderly and some special patients.For the scenarios with micro action and complex semantic,we design a lip-reading recognition and identity authentication system named as BackLip.We promote the perceived action to human lip movement when speaking.We perform user identification by analyzing the characteristics of human mouth movement when speaking passwords,and realize the instruction interaction by analyzing the input content.In this system,the users just need to read the password silently for successfully authentication,without the need to make a sound.Due to the uniqueness of everyone’s speech,the attacker can’t invade even when he knows the password,which greatly improves the security of the system.In addition,after analyzing and identifying the instruction content,humancomputer interaction can also be realized.The combination of identity authentication and instruction recognition can provide more comfortable personalized services.The Fl-score of user identification can reach 97.4%and the average accuracy of instructions recognition is 88.29%.We focus on the problem of human activity recognition and user identification based on passive backscattered signal.The results of our work show that both human behavior sensing and user identification can achieve high-precision accuracy in lightweight computing devices,with low-energy consumption and certain anti-interference ability.Therefore,out work is convenient to be deployed in IOT devices,providing new possibility for perception applications.