ChPOS: A Contactless and Continuous Method for Estimation of Heart Rate from Face

Recent years have seen an increase in interest in remote photoplethysmography (rPPG), as it is a fast, inexpensive, and convenient method for contactless estimation of a person's heart rate from facial videos and has potential in cardiac monitoring. Compared to traditional photoplethysmography (PPG), rPPG offers ease of access for disadvantaged and vulnerable members of the population, as this method saves cost and time by reducing frequent visits to the hospital. However, there are currently limitations to using rPPG in practice due to issues with consistent response skin colour, subject movement, and lighting artefacts. In this work we develop a new framework, ChPOS, by combining two traditional algorithms, CHROM and plane orthogonal to skin (POS). We modified the POS algorithm by incorporating the additional feature of the chrominance colour signal and changing the projection axis, to improve the accuracy of heart rate detection on subjects with darker skin complexion. The performance of our model is validated on two publicly available datasets, UBFC-RPPG and PURE. We compare the approach with state-of-the-art algorithms and the results show that our algorithm outperforms state of art models in the estimation of heart rate, with a mean absolute error (MAE) of 5.71 and root mean squared error (RMSE) of 7.27 on the UBFC-RPPG database and MAE of 5.39 and RMSE of 6.61 on the PURE database.