An Identity Authentication Method Based on Accelerometer and Gyroscope

In this paper, we propose a behavioral biometric authentication method based on accelerometer and gyroscope. The novelty of the proposed lies in feature extraction and similarity calculation. For feature extraction, a newly designed ConvBiGru-FCN network is employed to extract walking features from the multi-dimensional time-series acquired by the accelerometer and gyroscope. For similarity calculation, Tanimoto coefficient is used instead of conventional measures to calculate the distance between feature vectors. A dataset of 50 users has been collected in a realistic test ground for use in model training and testing. Extensive experiments have been carried out to evaluate the performance of the proposed scheme. Numerical results show that the proposed scheme with typical settings achieves an identity authentication accuracy of 93.10%.