Walk to Show Your Identity: Gait-based Seamless User Authentication Framework Using Deep Neural Network

With a rapid increase in the usage of wearable IoT devices such as a smartwatch, we can easily monitor various user activities which exhibit distinct patterns for each individual. Such activities as arm swings while walking orgait, can be used to distinctly identify different users. Therefore, this indirect interaction between the users and wearable IoT devices can be used as a biometric authentication technique to seamlessly authenticate and identify users. Thus, various gait-based authentication frameworks using sensor data collected through wearable or hand-held devices were proposed in the literature. However, many of them limitedly utilized the unique patterns by extracting features from the data sequences. Moreover, they require users to walk long period of time to collect large volume of sensor data in each authentication process, which hinders prompt user authentication. To address the limitations, we propose a gait-based seamless authentication framework using deep neural network (Deep Gait). Unlike many of the existing works, Deep Gait authenticates users without any feature extraction process, which can capture the overlooked features in the existing works. Moreover, Deep Gait requires less amount of sensor data (only one walk cycle) than the existing works (8 to 20 walk cycles) for user authentication, which enables seamless access control. Our experimental results evaluated on the commercial smartwatch show that Deep Gait achieves an Equal Error Rate (EER) of 1.8% which is lower than the existing works.