Heart rate measurement based on spatiotemporal features of facial key points

Heart rate monitoring among assembly line workers in workshops holds paramount significance for the proactive identification of safety hazards. To address the accuracy problem of remote photoplethysmography (rPPG) caused by motion artifacts and different light sources, we propose heart rate measurement based on spatiotemporal features of facial key points (HRMSF). Initially, to reduce the impact of motion artifacts on the region of interest, the spatial dimension features and the temporal dimension features of the key points in the region of interest are integrated, and a multipatch spatiotemporal feature (MSF) acquisition method for facial key points based on a unit time window is designed. Subsequently, to reduce the impact of different light sources on blood volume pulse (BVP) signal estimation, an extra green component (EGC) model based on the red and blue channels is constructed. Chroma and EGC are integrated into the pulse signal expression, resulting in the formulation of a new rPPG method named CHROM-EGC. Finally, PURE and ECG-Fitness are employed as test datasets, and the experimental results show that the MAEs obtained by HRMSF are 0.96 and 24.39, respectively, which are better than those of the 8 rPPG methods. Furthermore, MSF is migrated to 8 existing rPPG methods, and test results reveal that MSF can improve the performance of existing rPPG methods.