University of Southampton
From pixels to pulse: enhancing trust, quality and robustness in remote video-based pulse measurement
Abstract
dc:description.abstractRemote photoplethysmography (rPPG) enables non-contact heart rate measurement using everyday cameras, offering a promising alternative to traditional contact-based methods like electrocardiography and photoplethysmography. By leveraging subtle changes in skin color and micro-movements induced by blood flow, rPPG has the potential to revolutionize health monitoring. However, despite its potential, challenges such as motion artifacts, variations in lighting conditions and dataset biases challenge its robustness and reliability. This thesis investigates the foundations of rPPG, presenting a comprehensive study across signal processing, machine learning, video quality assessment and uncertainty quantification. We explore traditional signal processing techniques for rPPG, establishing a baseline for pulse estimation while highlighting their sensitivity to motion artifacts. To address these limitations, we propose a novel spatiotemporal two-stage learning framework (ST2S-rPPG), which integrates video stabilization, machine learning and adaptive region of interest selection to enhance pulse estimation accuracy. Recognizing the influence of video quality on rPPG performance, we systematically analyze the impact of motion, resolution, illumination and occlusions among other video quality factors, introducing video quality metrics tailored to rPPG. These metric provide a structured approach to assess video suitability for pulse extraction. Finally, we explore the application of conformal predictions to rPPG, establishing a framework for uncertainty quantification and compare MAE-based and quality-aware nonconformity measures. The findings of this thesis contribute toward making rPPG more practical for real-world deployment, with applications ranging from remote patient monitoring and telehealth to mental health assessments and human-computer interaction. While challenges remain in generalization across diverse populations and environmental conditions, these advancements lay a foundation for future research in making rPPG a reliable and scalable tool for healthcare and beyond.<br/>
Degree
thesis:*- Name dc:type.qualificationname
- Ph.D.
- Level dc:type.qualificationlevel
- doctoral
- Grantor dc:publisher.institution
- University of Southampton
- Year dc:date.issued
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Kateri, Eirini
- Advisors dc:contributor.advisor
-
- Farrahi, Kate
- Prugel-Bennett, Adam
- ORCID person:identifier.orcid
Rights
- Language dc:language
- English, eng