Facial Video-based Remote Photoplethysmography Signal Estimation with Vision Transformer

Remote photoplethysmography (rPPG), which is measured remotely with a video camera under ambient light conditions, is actively researched due to its accessibility and convenience. rPPG analysis can also estimate important biomarkers such as heart rate (HR) and heart rate variability (HRV). The majority of previous studies primarily designed algorithms to directly estimate HR or HRV. Instead, we note that regression analysis is used to recover the rPPG itself. This method has two benefits. Photoplethysmography (PPG) derived from a contact sensor provides crucial information for refining rPPG and insights for model development. Motivated by these considerations, we propose a model for end-to-end rPPG signal estimation. Vision transformer (ViT)-based model, which takes into account long-range correlations of rPPG, is adopted as the network's backbone, and the proposed model is evaluated using the MPSC-rPPG dataset.