Research On Key Technologies Of Intelligent Wireless Sensing Oriented To Human-computer Interaction

Gesture recognition and identity authentication technologies are the enablers of human-computer interaction(HCI).Compared with sensing technologies such as cameras and radars,although Wi-Fi-based wireless sensing technologies are limited by low bandwidth and low sensing resolution,they still have the advantages of low cost,easy deployment,unaffected by light and occlusion,and effectively controlling privacy leakage problem thus have received more and more attention.The wireless sensing technology empowered by artificial intelligence(AI),that is,intelligent wireless sensing(IWS),is widely regarded as a potential solution to realize ubiquitous sensing.This dissertation focuses on the research of key technologies of IWS for HCI,covering the construction method of pervasive intelligent wireless sensing system,cross-domain gesture recognition and identity authentication based on IWS,and the design solution of safe and efficient intelligent wireless sensing.The main innovations are as follows:1.Construction method of pervasive intelligent wireless sensing systemAt present,the research field of IWS lacks a universal sensing system construction standard and principle,which leads to the scattered and inefficient construction of IWS,and it is difficult to meet the increasing demand for ubiquitous intelligent sensing and intelligent services in the Internet of Things(IoT)era.To address these problems,this dissertation proposes a pervasive intelligent wireless sensing system architecture:covering the key points from sensing network deployment,sensing signal parameter feature selection,sensing signal processing and data set construction and intelligent sensing model construction,to adapt to the future new needs of intelligent wireless sensing scenes.This dissertation is a supplement and extension of existing research.2.Gesture recognition technology based on Wi-Fi sensing spatiotemporal characteristicsAiming at the congenital problem of insufficient spatial and temporal resolution of Wi-Fi sensing,the problem can be solved by effectively using multi-antenna and multi-carrier technologies to transfer the problem to high-discrimination sensing feature extraction,and then training a highly robust intelligent model.To this end,this dissertation proposes a Wi-Fi gesture recognition model based on 3D Convolutional Neural Network(3D-CNN),where 3D-CNN is used to obtain the spatio-temporal characteristics of wireless sensing data.At the same time,to adapt the sensing data to the three-dimensional model,this research cuts the sensing data in the three-dimensional space;because different people have different time to perform the same gesture,the time for the same person to perform different gestures will also vary,leading to that the data format of the sensing data is not uniform,that is,the data length is different.So this dissertation uses data filling to keep the size of the data block consistent in the third dimension of the sensing data block,and then adapt the threedimensional model.Experiments based on measured data show that the proposed model has good cross-domain recognition performance:Compared with the prior art,the proposed method has at least three percentage points improvement even if it crosses three domain factors at the same time and has similar performance when even crossing four domain factors at the same time;The model has the advantages of easy convergence,convenient training,and uncomplicated hierarchical structure.This research result can guide the design of a lightweight,highperformance gesture recognition model based on IWS.3.Identity authentication technology based on wireless characteristics of human bodyHCI has an urgent need for identity authentication.Current identity authentication based on wireless sensing usually relies on mining the wireless characteristics of the human body hidden in behaviors such as gait and breathing,that is,the separation of authentication and interaction.If identity authentication and human-computer interaction are seamlessly combined,the user experience will be greatly improved.To address the above problems,this paper proposes a gesture-based identity authentication scheme:tap the unique influence of individual gestures on the surrounding Wi-Fi signal propagation,and complete identity authentication while the user is trying to manipulate the machine.In addition,limited by the Wi-Fi sensing resolution,as a fine-grained behavior,the fluctuation of wireless signal propagation caused by gestures is limited,and it will be more difficult to extract the fluctuations of different users'gestures.In order to solve the above problems,this dissertation proposes a wireless authentication model based on the spatio-temporal attention mechanism:based on the extraction of the spatio-temporal high-resolution characteristics of the sensing data,it focuses on the difference and improves the performance of identity recognition.Experiments based on measured data show that the proposed model has an accuracy of 97.72%under the configuration of one transmitter and six receivers,and it has good cross-gesture robustness.This research has important theoretical and practical significance for guiding the design of an identity authentication system based on IWS.4.Joint gesture recognition and identity authentication technology under small sample conditionsCollecting and labeling sensing data is a labor-intensive task,which limits the development of intelligent wireless sensing technology to a certain extent.One of the commonly adopted solutions is small sample learning,that is,a small amount of labeled/unlabeled data is used to complete model training tasks that usually require a large amount of sample data.To this end,this paper proposes a dual-task learning model based on small samples:through in-depth research on the mechanism and internal connections of gesture recognition and identity authentication based on IWS,a dual-task learning model is designed to reuse sensing data sets;in particular,to further reduce the dependence on the number of samples,the relationship network is introduced into the dual-task model,and only a few new scene data are needed to achieve the recognition performance of the traditional method under the condition of large-scale data sets.Experiments based on measured data show that:?The proposed model can achieve high-precision gesture recognition and identity authentication performance under the condition of only one sample data of the new category.While effectively reducing the amount of sample data,it improves the model's ability to new categories.The generalization ability of,enhance the reusability of the model;?Cross-scene gesture recognition and identity authentication performance are better than current methods,identity authentication and gesture-based human-computer interaction are seamlessly connected,which enhances system security and improves user experience.This research is a strong support for the realization of ubiquitous intelligent sensing.5.Safe and efficient cross-domain intelligent wireless sensing methodWith the advent of the Internet of Things era,a large number of intelligent terminals that can communicate,calculate and sense have brought sufficient diversity of sensing data to the edge network,and also brought data privacy and security issues.With sufficient diverse sensing data and computing resources,it is expected to obtain an intelligent sensing model with excellent performance in specific scenarios.And we always expect that the model obtained by consuming a lot of resources can be suitable for new application scenarios,that is,the model has good crossdomain sensing ability,but in reality,the model is sensitive to the domain.How to make full use of the sensing data distributed in the edge network while taking into account the protection of data privacy is a difficult problem to be solved urgently.To this end,this article proposes an IWS solution powered by edge intelligence and blockchain:Edge intelligence provides sufficient low-latency computing and storage resources for IWS at the intelligent service demand side;blockchain helps protect users privacy-related sensing data and facilitate the transmission of sensing data across edge networks;cross-domain wireless sensing technology based on adversarial transfer learning efficiently utilizes computing,storage and data resources distributed in edge networks to provide low-latency,highly reliable intelligence on the service demand side service.Experiments based on measured data show that the cross-domain wireless sensing model based on adversarial transfer learning only needs 30%of the unlabeled data in the target scenario of the edge network to achieve the performance of the original model,effectively reducing the dependence of the smart model on labeled data and enhancing The reusability of the model is improved.The results have important theoretical and practical significance for guiding the design of efficient cross-domain IWS systems under privacy protection conditions.