Heart rate estimation based on face video under unstable illumination

Remote photoplethysmography (rPPG) is a noncontact heart rate (HR) measurement technique. The current heart rate measurement methods based on rPPG all require ideal lighting conditions, but the lighting in real scenes is complicated, Therefore, this article proposes a robust heart rate measurement method when unstable light (time-varying light and uneven spatial illumination) exists. First, the method locates the ROI(Region of Interest) area of ​​the face, divides it into blocks, and uses the color signals of different sub-blocks to establish a three-dimensional rPPG model. Second, the method performs logarithmic operations on each frame of the image to convert the relationship between the illumination component and the reflection component from a product to a sum so that the reflected component and noise can be separated in the frequency domain. Then, the ensemble empirical mode decomposition (EEMD) is used to decompose the reflected component, and the obtained intrinsic mode function (IMF) is applied to obtain the waveform reflecting the change in the heart rate. Finally, the signal quality (SQ) of each ROI sub-block is calculated, and the high-quality signals are combined to reconstruct the heart rate signal.