A data driven in-air-handwriting biometric authentication system

The gesture-based human-computer interface requires new user authentication technique because it does not have traditional input devices like keyboard and mouse. In this paper, we propose a new finger-gesture-based authentication method, where the in-air-handwriting of each user is captured by wearable inertial sensors. Our approach is featured with the utilization of both the content and the writing convention, which are proven to be essential for the user identification problem by the experiments. A support vector machine (SVM) classifier is built based on the features extracted from the hand motion signals. To quantitatively benchmark the proposed framework, we build a prototype system with a custom data glove device. The experiment result shows our system achieve a 0.1% equal error rate (EER) on a dataset containing 200 accounts that are created by 116 users. Compared to the existing gesture-based biometric authentication systems, the proposed method delivers a significant performance improvement.