Method of Selecting the Most Discriminatory Areas Based on Spectral Entropy in Remote Photoplethysmography

Abstract

A person's heart rate can be determined from a video image of the facial skin. This is explained by the fact that variations in the optical characteristics of biological tissues are caused by changes in their blood volume due to the pulse waves. Remote photoplethysmography problem is stated which is directed to estimation of a person's heart rate by registering of changes in volume pulse remotely, using a camera. Common drawback of the existing methods of the remote photoplethysmography is caused by facial image noise due to human mimic activity, head turns during video recording, non-stationarity of illumination and other similar factors. The main ways to solve such problems are to detect face in each frame of the video sequence, and to select the most informative regions of interest. A new method for selecting facial areas in a video is proposed, which leads to improve the accuracy of a solution of the remote photoplethysmography. The method consists in the application of criterion based on the Shannon spectral entropy to the color signals obtained from different facial areas on the video sequence in order to select the least noisy area. The correctness of the proposed method was confirmed in the course of experiments on an open database collected from 42 volunteers. The developed method reduced the relative error of heart rate estimation to 7 %.