Illumination variation interference suppression in remote PPG using PLS and MEMD

A novel framework based on partial least squares (PLS) and multivariate empirical mode decomposition (MEMD) is proposed to effectively evaluate heart rate from webcam videos captured during illumination changing conditions. The framework takes the assumption that both facial region of interest (ROI) and background ROI have the similar illumination variations and the background ROI can be treated as the denoising reference by using PLS to extract the underlying common illumination variation sources existing in both ROIs. Then, a number of intrinsic mode functions are decomposed by applying MEMD to the illumination-variation-suppressed facial ROI and the HR is evaluated. Compared to the experimental results obtained by the recently proposed independent component analysis and the ensemble empirical mode decomposition methods, the proposed method led to a better agreement with HR ground truth (the mean bias was 3.4 bpm with 95% limits of agreement ranging from −13.2 to 19.9 bpm), indicating a promising solution for the realistic HR estimation remotely.