Improved remote estimation of heart rate in face videos

We present an improved method for the remote estimation of heart rate from face videos. By relying on visual techniques used for the decomposition of spatial information, we propose a novel pipeline to properly extract and process the time signal containing the heart rate information. In order to validate our approach, a dataset of videos containing subjects in a stationary position has been recorded, together with the ground truth heart rate values measured with conventional ECG sensors. Experiments show the significant benefit of the proposed procedure with respect to existing techniques proposed in the literature.