Model-based sensor fusion of multimodal cardiorespiratory signals using an unscented Kalman filter

Abstract Based on a model of three coupled oscillators describing the influence of respiration, namely respiratory sinus arrhythmia (RSA), and so-called Mayer waves on the heart rate, an unscented Kalman filter (UKF) is designed to perform sensor fusion of multimodal cardiorespiratory sensor signals. The aim is to implicitly use redundancy between the sensor signals to improve the estimated heart rate while utilising model knowledge. The effectiveness of the approach is shown by estimations of the heart rate on synthesised data as well as patient data from the Fantasia dataset and a Sleep laboratory which provide two, three or six sensor channels for resting individuals. It could be shown that the approach is able to fuse multimodal sensor signals on signal level to achieve more accurate estimations. For real data, errors in mean heart rate as small as 1.56 % were achieved.