Remote SpO2 Estimation using End-to-End CNN Model

Blood oxygen speed (SpO2) is an indicator of the normal presence or absence of the respiratory function. This is attracting the attention of researchers since it could monitor the patient conditions of the chronic pulmonary diseases and covid-19. Covid-19 patients have the symptom of the significant SpO2 drop. This study tries to develop an early and easy checkup system of the continuous SpO2 using an RGB camera. Unlike the contact SpO2 measurement using the conventional optical sensor on a fingertip, the remote SpO2 sensor system is proposed using the facial video stream. The facial images are trained for the convolutional neural networks to implement the non-contact SpO2 estimation model, which is designed based on the architecture of the conventional remote photoplethysmography model.