Experiment with GMM-Based Subject Identification from PPG Signals Acquired by Wearable Sensors

This paper describes an experiment using the Gaussian mixture models (GMM) classifier for biometric identification from photoplethysmography (PPG) signals acquired by wearable sensors. The PPG-based identification method represents a simpler replacement of electrocardiogram-based authentication systems having several practical limitations. The performance of the open-set GMM-based identification system was tested depending on the number of Gaussian mixtures and the length of processed PPG signals. The obtained overall mean subject identification accuracy about 88 % is promising in this state of research. The performed first-step experiments confirm that the proposed conception of a subject identification system using the second-derivate PPG wave taken in conditions of low magnetic field with radiofrequency disturbance is functional. However, prior to practical usage as a real-time application, some implementation and optimization tasks will have to be solved.