Biometric Recognition based on Heart-Induced Chest Vibrations

The interest in wearable devices has tremendously increased in recent years, thanks the wide range of possible applications in which they can be used. As the employed sensors are able to capture relevant physiological signals, it has been lately proposed to exploit data collected by these instruments for automatic biometric recognition purposes. In this regard, a significant amount of studies has been conducted on the discriminative characteristics of the hearth activity. Within this context, in this paper we investigate the feasibility of recognizing people exploiting measurements of the mechanical vibrations of the chest, induced by the heart activity. To this aim, the effectiveness of combining signals acquired through seismocardiography and gyrocardiography is here evaluated. In order to evaluate the existence of discriminative characteristics within the considered signals, deep learning techniques relying on transfer learning have been applied on the considered measurements. Tests performed on ten subjects have been conducted in order to assess the influence on recognition performance of the positioning on the chest of the employed sensors, as well as the dependence on the posture assumed by the subjects while taking the employed recordings. The obtained results testify that promising recognition rates could be actually achieved when properly placing the considered devices.