Deep Super Resolution for Recovering Physiological Information From Videos

Daniel McDuff; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2018, pp. 1367-1374

Abstract


Imaging photoplethysmography (iPPG) allows for remote measurement of vital signs from the human skin. In some applications the skin region of interest may only occupy a small number of pixels (e.g., if an individual is a large distance from the imager.) We present a novel pipeline for iPPG using an image super-resolution preprocessing step that can reduce the mean absolute error in heart rate prediction by over 30%. Furthermore, deep learning-based image super-resolution outperforms standard interpolation methods. Our method can be used in conjunction with any existing iPPG algorithm to estimate physiological parameters. It is particularly promising for analysis of low resolution and spatially compressed videos, where otherwise the pulse signal would be too weak.

Related Material


[pdf]
[bibtex]
@InProceedings{McDuff_2018_CVPR_Workshops,
author = {McDuff, Daniel},
title = {Deep Super Resolution for Recovering Physiological Information From Videos},
booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2018}
}