Direct-Global Separation for Improved Imaging Photoplethysmography

Jaehee Park, Ashutosh Sabharwal, Ashok Veeraraghavan; The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2018, pp. 1375-1384

Abstract


Camera-based estimation of vital signs has made significant progress in last few years. Despite of the significant algorithmic advances, the low signal-to-background ratio in video-based photoplethysmography continues to be a performance bottleneck. One of the main challenges is that much of the light returning to the camera from the subject is surface reflection from the skin and other dermal layers, and hence does not contain any pulsatile blood perfusion information to estimate photoplesthysmogram (PPG). In this paper, we show that direct-global separation techniques designed to reject much of the surface reflection photons can improve the signal-to-background ratio in the raw captured video signal. We study two techniques for the suppression of direct surface reflection (a) cross-polarization and (b) structured illumination. Using a dataset from 28 participants, our results show an average SNR improvement in estimating PPG from the use of structured illumination is 1.42 dB compared to the brightfield illumination. The use of cross-polarizers leads to an average SNR increase of 1.49 dB compared to brightfield illumination. And the combined structured illumination and polarizer method increases the SNR on the average by 1.90 dB compared to the brightfield illumination. The key result is that local PPG estimate SNR can increase to more than 5.63dB, enabling very large gains on regions with large specular component. The RMSE decreased 55% and the range of error reduced by 12.9% with the use of polarizer and structured illumination.

Related Material


[pdf]
[bibtex]
@InProceedings{Park_2018_CVPR_Workshops,
author = {Park, Jaehee and Sabharwal, Ashutosh and Veeraraghavan, Ashok},
title = {Direct-Global Separation for Improved Imaging Photoplethysmography},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2018}
}