Research On Wireless Signal Based Cross-site And Large-scale Activity Recognition And User Privacy Information Protection

Wireless signal basd sensing technologies are emerging as a vital component of Internet of Things,and has received much attention in academia and industry in recent years,because that its deployment is simple,it need no support of additional hardwares,it performs well beyond line-of-sight,it has no light limitation,and so on.Channel State Information(CSI)-based sensing technology has become part of mainstream technologies because of high accuracy and rich signal information,and it has important applications in smart home,medical monitoring and etc.However,there still exist problems in pratical applications.First,in case of large-scale sensing,the accuracy will be low because of the similarity between wireless signals of different activities,which restricts the large-scale deployment of system.Second,in case of cross-site sensing,the cost will be high because the characteristic modes of the same activity will be totally different across scenarios and current methods need to collect training data from each scenario,which is a repetitive work and will lead to the waste of human resources.Third,with the development of sensing technology and popularity of wireless devices,security issues arise when the system is applied in pratical scenarios.User privacy information,such as pattern unlock patterns and digital payment passwords,may be leaked.Considering CSI-based sensing technology,this paper analyzes the problems of largescale sensing,cross-site sensing and user privacy information protection in pratical applications.Aiming at high accuracy,low cost,strong robustness and security,this paper proposes two sensing methods,which separately solve large-cale and cross-site sensing problems,two methods to protect user privacy information.The main contributions of this paper are as follows:(i)This paper proposes a large-scale activity recognition method.Mixture-of-expert is leveraged in this paper,and the basic idea is that there always exists one model to successfully identify the testing sample.Specifically,a model selector is trained by previous collected training samples,and when given a testing sample,the model selector will select a suitable model for the testing sample.Experimental results show that our approach not only improves the accuracy compared to current methods,but also has strong stability,i.e,the accuracy will not decrease with the increase of scale.(ii)This paper proposes a cross-site activity recognition method.Artificial neural network is used to train a roaming model in this paper.Specifically,a roaming model is first trained using the training samples from two scenarios,and then the roaming model can transform the data in the source scenarios to the data in the target scenarios,and the transformed virtual data can be seen as the fingerprints in the target scenarios,without requiring the user to recollect the training samples in target scenarios.Experimental results show that our approach reduces the cost of collecting training samples in multiple scenarios while guaranteeing the high accuracy.(iii)This paper proposes a channel interference based user privacy information protection method.The central idea is to destroy the necessary requirements of a successful CSI-based attack,and we leverage channel interference to protect user's privacy information by interfering the attacker's wireless devices.Specifically,after a user enters the public environment,the system continuously monitors whether there is a CSI-based attack in the environment by analyzing the network activity,and when an attack is detected,the channel interference protection method will be activated and the system will use a normal wireless device to interfere the attacker's wireless devices.Experimental results show that our approach can defeat CSI-based attacks.(iv)This paper presents a user privacy information protection method based on safe region guidance.Through the analysis,we know that a successful CSI-based attack need the user be close to the attacker's wireless devices,and we guide a user to a safe region to protect user privacy information.Specifically,the mobile phone continuously collect the sensors data during the user's walk,and after detecting a CSI-based attack,the mobile phone will locate the attacker's wireless devices using current developed localization technology,and combined with the localization results,the system will guide the user to a safe region.Experimental results show that our approach can successfully defeat CSI-based attack.