Fast Skin Segmentation on Low-Resolution Grayscale Images for Remote PhotoPlethysmoGraphy

Abstract

Facial skin segmentation is an important preliminary task in many applications, including remote PhotoPlethysmoGraphy (rPPG), which is the problem of estimating the heart activity of a subject just by analyzing a video of their face. By selecting all the subject’s skin surface, a more robust pulse signal could be extracted and analyzed in order to provide an accurate heart activity monitoring. Single-photon avalanche diode cameras have proven to be able to achieve better results in rPPG than traditional cameras. Although this kind of cameras produces accurate photon counts at high frame rate, they are able to capture just grayscale low resolution images. For this reason, in this work, we propose a novel skin segmentation method based on deep learning that is able to precisely localize skin pixels inside a low-resolution grayscale image. Moreover, since the proposed method makes use of depthwise separable convolutional layers, it could achieve real-time performances even when implemented on a small low-powered IoT device.