Indoor Device-Free Passive Intrusion Detection Based On Channel State Information

With Internet of things(IoT)coming into people’s everyday life,people’s life becomes even smarter day by day,and smart home techniques are being accepted by more and more people.More specifically,the indoor intrusion detection system is one of the most important component of smart home.The novel wireless signal-based device-free passive intrusion detection systems do not require users to carry dedicated electronic devices.It only utilizes the wireless signal in the environment to detect if there is someone in the area of interest,or even identify his identity.Recently,with the popularization of WLAN techniques,the researchers find that beyond communication,the wireless network can also be used as the wireless sensor network to sense the environment.The IEEE 802.11 n network uses OFDM and MIMO techniques,which we can utilize to obtain finer-grained subcarrier-level channel respond.It gives a good opportunity for us to implement indoor intrusion detection using WiFi signals.Although large quantities of researches have been done on the sensing capabilities of WiFi signals,there are still many application problems in passive intrusion detection based on WiFi signals,including the robustness of human detection and the effectiveness of identity identification.In this thesis,we dig deep into the fine-grained channel state information,and utilize it to detect human and identify human identity more reliablely.It will make CSI-based indoor intrusion detection more effective.In this thesis,we focus on indoor intrusion detection,and mainly conduct the research from the following aspects:Firstly,in order to improve the human detection performance when the intruder moves very slowly,we propose a moving speed independent passive human detection approach.It utilizes a novel feature,which can extract the fluctuation of the whole channel.The feature is more sensitive to the changes of the environment.In addition,we transfer the human detection problem into a probability problem,which makes the system have the ability to detect human of different moving speeds.We conduct numerous experiments in this thesis,and compare the approach with other methods.Compared to the current methods that use only a single subcarrier,the approach proposed in this thesis can improve the detection accuracy when people move very slowly without increasing the computational complex.The detection false positive and false negative is lower than 1%,which is a satisfactory detection performance.Secondly,to deal with the problem that human detection systems will fail to detection the intruder when the intruder moves unconventionally in indoor environment.This thesis proposes a robust moving pattern independent passive human detection approach to extract the robust frequency-domain feature using continuous wavelet transform from all subcarriers.The frequency-domain feature can sense the subtle changes in the channel.It can be adapted to different indoor scenarios.Compared to the current human detection methods,the approach in this thesis can reduce the cost of site-survey,and can also detect intruder who moves unconventionally.The experiments indicate that it can achieve a FN and FP rate of 2% under different moving patterns,which is effective in intrusion detection.Thirdly,the intrusion detection system has the requirement to recognize the stranger,but the current human identification approaches do not have the capability to do it.Usually the strangers are not in the training set,which is a big challenge for a typical classifier.In this thesis,we propose a stranger recognition approach,using time-frequency transform techniques,and it extracts frequency-domain gait features from channel state information to build two Gaussian Models of authenticated persons and strangers,respectively.The Gaussian Mixture Model acts as the classifier.We conduct experiments using the walking data of 6 people in two typical rooms.The average FN and FP achieves 15% and 12%,respectively,which indicates the effectiveness of the approach.Finally,in order to improve the identification accuracy of human identities,in this thesis,we propose an indoor passive identity identification approach based on gait time–frequency analysis.We utilize time–frequency analysis techniques to divide the time series into step segments and walking segments,and extract step features and walking features.The most representative features are selected based on information gain to reduce computation complex.Numerous expriments are conducted that the approach can achieve the identification accuracy of 98.7% among two people,while 90.9% among eight people,which indicates the effectiveness of device-free passive identity identification using channel state information.