Device-Free and Robust User Identification in Smart Environment Using WiFi Signal

摘要

There has been a booming interest to build the smart indoor environment solution such as smart home or office, which is capable of sensing and responding to the users in a device-free manner. Among all the enabling technology of such smart environment solution, WiFi-signal based device-free sensing has shown great potential for its easy and low-cost deployment and various sensing functions. In such environment, identification is the key component of the environment, which should only respond to legitimate users. However, the state-of-theart work only uses the gait information to perform identification, which suffers from two problems. First, the identification process requires a relatively long walking to gather enough information, which restricts such method to a large open indoor environment. Second, some functions of the smart environment should be triggered pro-actively by the user specified activities, while the gait-based solution is not flexible enough to do so. To solve above problems, we propose WiID, which innovates in using channel state information (CSI) of commercial WiFi devices to perform identification by extracting information from humans' general activities. Comparing to the other existing solutions, such scheme provides the advantages in efficiency, scalability, and reliability. Specifically, we first propose an activity-based identification model, which considers both explicit and implicit information of the human activities to describe the relationship between user's identity and activity. We then design several algorithms to robustly extract and recognize above information from the CSI of WiFi signal. Finally, the extracted information is used to perform identification and proved to have the potential to resist shoulder surfing attack. Based on above model and algorithms, we implement WiID on commercial WiFi devices and evaluate it in two different environments. Our results show that WiID achieves an average accuracy greater than 90% with only a small number of combinations, which imply a great potential in dealing with the secure interaction and authentication problem in the smart environment.