| With the popularization of 802.11 n protocol,using ubiquitous and fine-grained WiFi signals to achieve human authentication has been possible.Different from traditional methods,which require extra devices(e.g.,cameras and ultrasonic devices),WiFi-based authentication can provide a device-free solution ubiquitously only using the Commercial-Off-The-Shelf(COTS)WiFi devices.Moreover,it owns unique advantages such as not requiring a line-ofsight path and preserving human privacy.Hence,lots of scholars have proposed identification schemes based on WiFi signals.However,these state-of-the-art methods still suffer from multiple disadvantages,which are as follows:(1)Segmenting activities based on experience makes it difficult to deploy the system;(2)Low accuracy for illegal recognition limits the system security;(3)Long recognition delay leads to poor user experience;(4)Unsuitable for environmental dynamics.To tackle all problems as mentioned above and develop a device-free yet accurate WiFi authentication system,in this paper,we design WiAU,which not only recognize human identities and activities by utilizing the Deep Learning technique,but also has transfer learning ability.The core model of WiAU consists of three well-designed components: preprocessing module,human recognition module,and transfer learning module.In the preprocessing module,to automatically segment human activities and extract human features,we designed the ASA(Automatic Segmentation Algorithm).In the human recognition module,to validate illegal users,authenticate legal users,and recognize activities,a deep learning model based on residual network(ResNet)and convolutional neural network(CNN)with two dedicated loss functions is designed.Moreover,based on the short-cut connection design of the ResNet,WiAU is able to recognize human with a short recognition delay.Furthermore,in the transfer learning module,we designed a transfer learning based algorithm and human height estimation algorithm against environmental dynamics.Finally,experiments are conducted to validate the performance of WiAU.Compared with the state-of-the-art device-free authentication systems,WiAU achieves promising accuracy over 98% and 92% in human authentication and activity recognition with a good transfer learning ability,respectively. |
