Noise-Assessment-Based Screening Method for Remote Photoplethysmography Estimation

Abstract

Remote vital signal estimation has been researched for several years. There are numerous studies on rPPG, which utilizes cameras to detect cardiovascular activity. Most of the research has concentrated on obtaining rPPG from a complete video. However, excessive movement or changes in lighting can cause noise, and it will inevitably lead to a reduction in the quality of the obtained signal. Moreover, since rPPG measures minor changes that occur in the blood flow of an image due to variations in heart rate, it becomes challenging to capture in a noisy image, as the impact of noise is larger than the change caused by the heart rate. Using such segments in a video can cause a decrease in overall performance, but it can only be remedied through data pre-processing. In this study, we propose a screening technique that removes excessively noisy video segments as input and only uses signals obtained from reliable segments. Using this method, we were able to boost the performance of the current rPPG algorithm from 50.43% to 62.27% based on PTE6. Our screening technique can be easily applied to any existing rPPG prediction model and it can improve the reliability of the output in all cases.