Real Time Gait based Person Authentication using Deep Hybrid Network

Gait is one of the behavioral biometrics which has a profound value in-person identification. Authentication by gait identifies a person with his/her unique walking pattern. Gait has the advantage of being unobtrusive compared to other biometrics. Gait authentication is important in various security, medical- and sports-based applications. This paper mainly analyses gait features depending on acceleration sensors of smartphones and proposes a new gait authentication method based on the late fusion technique. By using Deep Hybrid Network (DHN), we develop an effective model for real-time authentication of users. DHN is a fusion of convolutional neural network (CNN), long short term memory (LSTM), and Gated recurrent unit (GRU), which are sequentially added to the model, and their features are combined to make a final decision by a softmax layer. The combination of networks is made by spreading the input features and changing the neural network's initial weights. The performance of the proposed model is evaluated and compared with that of independent deep learning models with three benchmark datasets. The proposed model Deep Hybrid Network produces better authentication accuracy compared to individual models. It can be considered for user authentication in real-time.