Incorporating physiological dynamics in synthetic face video

Abstract

Remote photoplethysmography enables non-contact Heart Rate(HR) assessment by studying skin color fluctuations captured in facial videos. However, the availability of real face video datasets with physiological ground truth is severely limited due to privacy concerns and data collection challenges. In response, this thesis presents a novel method for integrating physiological dynamics into synthetic face videos generated through advanced synthesis techniques, specifically designed for benchmarking rPPG algorithms in healthcare and telemedicine applications. By integrating a controllable rPPG signal and simulating physiological skin tone variations, we create synthetic videos that are able to reflect the target HR. This method ensures both visual and physiological realism, offering a valuable resource for the development and evaluation of rPPG methods. Our results demonstrate that the embedded physiological signals can be reliably extracted from the generated videos, validating the effectiveness of this synthetic dataset for advancing rPPG research.