Oxygen Saturation Estimation Based on Optimal Band Selection from Multi-band Video

In this study, we propose a method to estimate oxygen saturation by selecting the best bands from video images captured by a multiband camera. Oxygen saturation is one of the most important bioindicators for measuring human health. For example, when a person contracts COVID-19, which is currently prevalent, oxygen uptake does not work properly and oxygen saturation drops without the person being aware of it, which may lead to severe symptoms. Monitoring oxygen saturation is very important so that the person receives treatment before such a situation occurs. The commonly used contact sensor is uncomfortable because of its pressure and it is difficult to wear on a daily basis, so non-contact estimation of oxygen saturation is desirable. To estimate oxygen saturation using a contact sensor, the difference in the absorption coefficients of oxidized hemoglobin and deoxidized hemoglobin is used. Using the same principle, it is possible to estimate oxygen saturation without contact using the signals from two channels obtained by an RGB camera. Currently, many smartphones are equipped with infrared cameras for face recognition, and increasingly more models are equipped with multi-camera systems consisting of RGB and infrared cameras. In such cases, it is difficult to take advantage of the multiple bands because the optimal combination of bands for oxygen saturation estimation varies depending on the imaging environment and the subject. In this study, to select the optimal combination of bands from multi-band video images, we used a Monte Carlo simulation of light scattering on the skin to simulate pulse waves during oxygen saturation changes while measuring the signals with a multi-band camera. We further propose a method to select the most accurate combination for estimating the oxygen saturation based on the features obtained from the pulse wave.