Probabilistic Signal Quality Metric for Reduced Complexity Unsupervised Remote Photoplethysmography

Remote photoplethysmography (rPPG) is a recent technique for estimating heart rate by analyzing the pulsatility of skin hue using regular cameras. To determine the quality of the measurement, many existing methods are based on the signal-to-noise ratio (SNR) calculated in the frequency domain. However, the Fast Fourier Transform (FFT) operation is performed with a minimal complexity of O(nlogn). Therefore, the use of this quality metric in an unsupervised rPPG framework in which this metric is estimated a large number of times will tend to greatly increase the complexity of the solution. In this paper, we propose a new probabilistic formulation of a cardiac signal quality index, with lower complexity, based on the Bayesian information criterion (BIC) that encapsulates the characteristic shape of the rPPG signal. The results of this study, obtained on a public database, have demonstrated that the proposed probabilistic metric outperforms the regular SNR metric with a lower computation complexity.