Ballistocardiogram Based Person Identification and Authentication Using Recurrent Neural Networks

Abstract

Physiological signals such as electrocardiogram (ECG) have been used as biometric in person identification and authentication during last decade. Ballistocardiogram (BCG) records the weight variation caused by cardiac contraction and ejection of blood, which shows individual variation and has potential in applications of biometric. In this paper, we investigated the possibility of BCG based subject identification by using recurrent neural networks (RNNs). Different RNN structures including simple RNN cell, long-short-term memory (LSTM) and gated recurrent unit (GRU) were evaluated. BCG, ECG and BCG-ECG union were used as the biometric and their performance were compared under the metric of identification accuracy and equal error rate (ERR). By the using BCG, the identification accuracy was 97.8 % for identification and the ERR was 0.9% for authentication. The accuracy and ERR of ECG were 98.9% and 0.5% respectively. The result achieved by BCG- ECG union was that the identification accuracy was 100% and the ERR was 0%. The result proves BCG carries individual information and can be used as a biometric for person identification and authentication. Additionally, multi cardiac signals union such as BCG-ECG union may have huge potential in person identification.