Adaptive Heart Rate Estimation from Face Videos

TL;DR

This paper introduces an adaptive method for estimating heart rate from facial videos by distinguishing and mitigating different noise types, significantly improving accuracy over existing approaches.

Contribution

It presents a novel adaptive HR estimation technique that separately handles rigid head motion and non-rigid facial expressions, enhancing robustness and accuracy.

Findings

Reduces HR estimation error by over 32%

Effectively distinguishes between noise types for targeted noise removal

Improves robustness of HR estimation in facial videos

Abstract

We propose a novel heart rate (HR) estimation method from facial videos that dynamically adapts the HR pulse extraction algorithm to separately deal with noise from 'rigid' head motion and 'non-rigid' facial expression. We first identify the noise type, based on which, we apply specific noise removal steps. Experiments performed on popular database show that the proposed method reduces HR estimation error by over 32%.