Head pose‐free gaze estimation using domain adaptation

Abstract

Human gaze information has been widely used in various areas, such as medical diagnosis and human-computer interactions (HCI). This study proposes a head pose-free 3D gaze estimation method using a deep convolutional neural network (DCNN). To infer gaze direction, only a small grayscale image is required without any special devices such as an infrared (IR) illuminator and RGBD sensor. A domain adaptation method to reduce the feature gap between real and synthetic image data is also proposed here. Moreover, a novel synthetic dataset (SynFace) that contains head poses, gaze directions, and facial landmarks is established and released. The proposed method outperforms state-of-the-art methods and achieves a mean error of less than 4 • .