Temporal-spatial smoothing mask for enhanced remote heart rate estimation

Remote photoplethysmography (rPPG), which estimates heart rate from facial video signals captured by inexpensive RGB cameras, has been shown to be effective at estimating heart rate in stable scenarios. Region of Interest (ROI) selection is a crucial aspect of mainstream rPPG heart rate estimation methods, as improper settings can introduce significant noise. In this paper, we reconsider ROI selection strategies from the perspective of mask smoothness and propose an ROI selection strategy called Temporal-spatial smoothing mask. This strategy first selects masks using the proposed YCbCr-LUT for spatial smoothing, and then temporally smooths the masks using Gaussian filtering. Experiments on the publicly available datasets UBFC-rPPG show that the proposed method effectively improves heart rate estimation performance.