Ballistocardiogram-based Authentication using Convolutional Neural Networks
It proposes a biometric authentication method for wearable device users, including those with disabilities, but is incremental as it applies existing CNN techniques to a new signal type.
This paper tackles authentication using ballistocardiogram (BCG) signals from head-mounted wearables, achieving an equal error rate (EER) of 3.5% for general subjects and 11.2% for those with motor disabilities immediately after training, though performance degrades over time.
The goal of this work is to demonstrate the use of the ballistocardiogram (BCG) signal, derived using head-mounted wearable devices, as a viable biometric for authentication. The BCG signal is the measure of an person's body acceleration as a result of the heart's ejection of blood. It is a characterization of the cardiac cycle and can be derived non-invasively from the measurement of subtle movements of a person's extremities. In this paper, we use several versions of the BCG signal, derived from accelerometer and gyroscope sensors on a Smart Eyewear (SEW) device, for authentication. The derived BCG signals are used to train a convolutional neural network (CNN) as an authentication model, which is personalized for each subject. We evaluate our authentication models using data from 12 subjects and show that our approach has an equal error rate (EER) of 3.5% immediately after training and 13\% after about 2 months, in the worst case. We also explore the use of our authentication approach for people with motor disabilities. Our analysis using a separate dataset of 6 subjects with non-spastic cerebral palsy shows an EER of 11.2% immediately after training and 21.6% after about 2 months, in the worst-case.