Gaze direction estimation using only a depth camera

Abstract

This paper proposed a novel method to perform gaze direction estimation from human eye movement with only a consumer level depth sensor (RealSense). Proposed method consists of four stages: (i) eye region localization from a template matching of the contour of the chin and geometric properties of human faces, (ii) locating the eye center points as the reference points, (iii) using second order Laplacian detector to strengthen the intensity of depth data to locate the pupil and (iv) gaze direction classification using the Euclidean distance and visual angle. Experiments are performed to estimate the performance of the gaze directions on low illumination environments and different testing distances between the camera and users. Finally, the experimental results demonstrate the proposed system using solely the distance information for gaze direction estimation.