Optimal facial regions for remote heart rate measurement during physical and cognitive activities

Abstract

Remote photoplethysmography (rPPG) has gained prominence as a non-contact and real-time technology for heart rate monitoring. A critical factor in rPPG’s accuracy is the selection of regions of interest (ROI), as it can significantly influence prediction outcomes. Most studies typically use the forehead and cheeks as ROIs, but little research has explored other facial regions or how stable these ROIs are during physical movement and cognitive tasks. In this study, we analyzed 28 facial regions based on anatomical definitions using two mixed datasets derived from three public databases: LGI-PPGI, UBFC-rPPG, and UBFC-Phys. We applied rPPG algorithms such as orthogonal matrix image transformation (OMIT), plane-orthogonal-to-skin (POS), chrominance-based (CHROM), and local group invariance (LGI). Our findings show that the glabella, medial forehead, lateral forehead, malars, and upper nasal dorsum consistently perform well, with the glabella achieving the highest overall evaluation score. These results offer valuable insights for advancing remote heart rate monitoring technologies.