Automatic region-based heart rate measurement using remote photoplethysmography.

This paper presents a model-based approach to measure the vital signs from RGB video files focusing on the heart rate. We use the plane-orthogonal-to-skin (POS) remote photoplethysmography (rPPG) transformation performed individually at five well-defined regions of interest (ROI) in the face. We extract the heart rate information by a correlation of the different rPPG signals in these ROIs and a magnitude-based reliability calculation. This increases the robustness of the heart rate extraction from videos. With this method, we achieve a mean of all calculated mean-absolute-errors of 8.324 BPM in the V4V-Challenge data (averaged over all videos of the training and validation set).