Non-Contact HR Extraction from Different Color Spaces Using RGB Camera

Nowadays, non-contact vital sign measurement from facial videos using an RGB camera has gained popularity among researchers as it is a feasible and convenient method suitable for personalized and clinical health monitoring. This paper proposes a simple but cogent technique for heart rate (HR) estimation from the facial RGB videos. It is suggested that the integration of color channels from different color spaces derived from the RGB model can provide a better estimation of the pulsating component of arterial blood synchronous with the cardiac cycle. The shared pulse signal related to blood volumetric changes underneath the skin existing in these color signals is separated using the principal component analysis, and the resultant signal is used to determine the HR value using the short-time Fourier transform. The experiments are performed using three publicly available datasets including PURE, UBFC-rPPG, and Cohface. In the experimental analysis, the proposed technique yields lower values of the mean absolute error (MAE) and root mean square error (RMSE) for the three datasets as, PURE: MAE = 1.65 beats per minute (bpm) and RMSE = 2.9 bpm, UBFC-rPPG: MAE = 2.57 and RMSE = 5.57 bpm, and Cohface: MAE = 4.51bpm and RMSE = 6.5 bpm. These performance measures for the proposed technique are found to be lower than those obtained from some of the state-of-art methods. This study suggests that color channels from the alternative color spaces can be used for non-contact vital sign monitoring.