Personal Identification by Human Motion Using Smartphone

Abstract

There are two types of personal authentication methods, that is, biometric authentication method and password authentication method. Although biometric authentication has an advantage that security is higher than password authentication, there is also a disadvantage that biometric information can not be replaced. On the other hand, an authentication method using human motion has been proposed in an existing research. Human motion is a kind of changeable biometric information even when biological information leaks out. In the authentication method, biometric information for authentication depends on an acceleration sensor and a gyro sensor. DP matching is used for the comparison of users in the research. However, false acceptance rate is 10.71% and false rejection rate is 56.90% in a single action. There is no way to use this method for personal identification. In our research, we also obtained data from acceleration sensors and gyroscopes which are standard equipments on usual smartphones. We used convolutional neural networks for the comparison of individuals instead of DP matching. As a result of personal identification by convolutional neural network with data of “circle”, “rectangle” and “check” which are acquired from examinees, individuals can be identified with a rate of more than 88.5% on average in multi-level classification and 78.2% on average in binary classification.