Deep face segmentation for improved heart and respiratory rate estimation from videos

Abstract The selection of a suitable region interest (ROI) is great importance in camera-based vital signs estimation, as it represents the first step processing pipeline. Since all further relies on quality signal extracted from ROI, tracking this area decisive for performance overall algorithm. To overcome limitations classical approaches such partial occlusions or illumination variations, custom neural network pixel-precise face segmentation called FaSeNet was developed. It achieves better results two datasets compared to state-of-the-art architectures while maintaining high execution efficiency. Furthermore, Matthews Correlation Coefficient proposed loss function providing fitting weights than commonly applied losses field multi-class segmentation. In an extensive evaluation with variety algorithms our able achieve both heart and respiratory rate estimation. Thus, ROI estimation could be created that superior other approaches.