Advances in Non-Contact Cardiovascular Monitoring: A Systematic Review of Face-Based Heart Rate and Oxygen Saturation Estimation With Machine Learning

Non-contact cardiovascular monitoring has gained significant attention due to its potential for seamless and continuous health assessment. Traditional methods for measuring heart rate (HR) and oxygen saturation (SpO) require direct contact with the skin, which can be inconvenient and unsuitable for certain applications. Recent advancements in computer vision and machine learning have enabled accurate estimation of these vital signs using facial videos. This paper provides a systematic review of face-based HR and SpO estimation techniques, emphasizing machine learning approaches. We discuss various methodologies, data sources, pre-processing techniques, and model architectures used in the field. Additionally, we compare the performance of different algorithms and highlight challenges such as motion artifacts, lighting variations, and model generalization. The review aims to present a comprehensive understanding of the current state of non-contact cardiovascular monitoring and identify potential future directions for improving accuracy and robustness.