A Comparative Analysis on the Impact of Face Tracker and Skin Segmentation onto Improving the Performance of Real-Time Remote Photoplethysmography

Abstract

AbstractRemote photoplethysmography (rPPG) is well known means for measuring heart rate in remote, by analyzing physical changes such as skin color caused by cardiac activity. Previous rPPG studies usually focused on the method that how to accurately extract rPPG signals from video clips using the fixed cameras in the stable environment. In this paper, we conducted a comparative analysis using the deep learning-based object tracker and the skin segmentation method, which are image preprocessing techniques to extract rPPG signals from facial video. Experiment results showed that the noise problem caused by the unstable tracking trajectory and bounding box coordinate of tracker can be solved by skin segmentation. In addition, the skin segmentation that defines the skin color space in the YCbCr and HSV color models did not affect a real-time rPPG processing speed, and rPPG accuracy was significantly improved. In conclusion, we experimentally verified that faster rPPG measurements without loss of accuracy are possible using faster image preprocessing algorithms based on observations found through comparative analysis of skin segmentation and object trackers for real-time rPPG.KeywordsRemote photoplethysmographyBiomedical monitoringRemote sensingHeart rate measurement