چكيده لاتين
Remote photoplethysmography (rPPG) is a non-contact method for measuring vital signs such as heart rate, blood oxygen levels, and respiratory rate by analyzing variations in light reflected from the skin, captured using video cameras. This technology utilizes signals related to subtle changes in blood volume beneath the skin caused by cardiac pulses to extract physiological parameters. Its applications include medical environments such as neonatal intensive care units (NICU), elderly care, and monitoring viral conditions like COVID-19. Using standard cameras, such as those on smartphones or webcams, rPPG can continuously and in real-time provide accurate information about an individual’s health status.
System Design Framework: This study focuses on designing a system for extracting vital signs from video using remote photoplethysmography (rPPG). The system includes stages for video recording, facial and region of interest (ROI) detection, rPPG signal extraction, and signal processing to calculate physiological parameters such as heart rate, blood oxygen levels, respiratory rate, and heart rate variability. Videos are divided into individual frames, and RGB cameras are used for data recording. Signals are extracted through various methods after noise removal. For more precise analysis, a windowing approach has been employed in this study.
Datasets: A dataset comprising videos of 81 individuals (65 males and 16 females), recorded in this project, was used to train and evaluate the system. Additionally, the UBFC-rPPG and UBFC-Phys datasets were also utilized.
In this study, various rPPG signal extraction algorithms, including CHROM, PBV, POS, ICA, LGI, and OMIT, were evaluated. The results indicate that the CHROM and POS algorithms demonstrated the best performance. The CHROM algorithm achieved an R² value of 0.91 and a CCC value of 0.94, while the POS algorithm achieved an R² value of 0.94 and a CCC value of 0.96, showing high accuracy in heart rate estimation. Additionally, Bland-Altman plots showed negligible differences between actual and estimated values, and the Pvalue for both algorithms was 0.2782, confirming no significant differences. These findings highlight the superiority of the CHROM and POS algorithms in accurately evaluating rPPG signal extraction.
