Search for: All records

In recent years, people with upper extremity impairment (UEI) have been using wearable Internet of Things (wIoT) devices like head-mounted devices (HMDs) for a variety of purposes such as rehabilitation, assistive technology, and gaming. Often such wIoT devices collect and display sensitive information such as information related to medical care and rehabilitation. It is therefore crucial that HMDs can authenticate the person wearing them so that appropriate access control can be implemented for the sensitive information they manage. In this paper, we explore a new authentication approach for people with upper extremity impairment (UEI) for wIoT devices head-mounted devices (HMDs). The approach works by leveraging ballistocardiograms - representations of the cardiac rhythm - derived from an accelerometer and a gyroscope, mounted on an HMD for authentication. The derived ballistocardiograms are then fed into six participant-specific convolutional neural networks (CNNs) which act as our authentication models. Analysis of our approach shows its viability. Using data from 6 participants with UEI (and 22 able-bodied participants, for evaluation), we show that we can authenticate a participant in 4 seconds with an average equal error rate of 4.02% and 10.02%, immediately after training and ~2 months later, respectively.