Person Identification Based on Accelerations on Drawing Figures with a Smartphone

Several methods to estimate the user who is holding a smartphone by analyzing the acceleration obtained from the smartphone's accelerometer using deep learning have been proposed. However, these methods have some issues such as insufficient accuracy or the need for the user to hold a smartphone for a long time. In this paper, we discuss the estimation of the user based on acceleration measured in a shorter aperiod of time. We propose a method to identify a user by make a user draw a figure in the air. The proposed method is based on the assumption that a user is estimated from users given in advance. Acceleration data of all users is acquired in advance, and learning is performed by deep learning using these acceleration data to create a model for estimation. The acceleration data measured for identification are analyzed using this model, and the user who is holding the smartphone is idenfitied. We evaluated the proposed method using two networks, LSTM and DeepConvLSTM, and showed that the proposed method can identify the user with high accuracy. In particular, the accuracy of the method using DeepConvLSTM is high.