The Comparison of Neural Networks for Contactless BPM Estimation

The quest for non-invasive vital sign monitoring has led to transformative advancements in healthcare technology. Among these, Remote Photoplethysmography (rPPG) represents a significant breakthrough, enabling the extraction of vital signs such as heart rate variability (HRV) and blood volume variations without physical contact. This study delves into extracting Photoplethysmogram (PPG) signals from images, probing the feasibility of computing Beats Per Minute (BPM) through various neural network architectures. Our experiments, conducted on a dataset comprising 18,000 images, show promising outcomes. Specifically, ResNet101V2 and DenseNet201 emerge as frontrunners, demonstrating superior performance in our experiments. Integrating deep learning techniques holds great promise for enhancing vital sign extraction in diverse healthcare applications.