A Geometry-Appearance-Based Pupil Detection Method for Near-Infrared Head-Mounted Cameras

Abstract

This paper presents, for the first time, a novel pupil detection method for near-infrared head-mounted cameras, which relies not only on image appearance to pursue the shape and gradient variation of the pupil contour, but also on structure principle to explore the mechanism of pupil projection. There are three main characteristics in the proposed method. First, in order to complement the pupil projection information, an eyeball center calibration method is proposed to build an eye model. Second, by utilizing the deformation model of pupils under head-mounted cameras and the edge gradients of a circular pattern, we find the best fitting ellipse describing the pupil boundary. Third, an eye-model-based pupil fitting algorithm with only three parameters is proposed to fine-tune the final pupil contour. Consequently, the proposed method extracts the geometry-appearance information, effectively boosting the performance of pupil detection. Experimental results show that this method outperforms the state-of-the-art ones. On a widely used public database (LPW), our method achieves 72.62% in terms of detection rate up to an error of five pixels, which is superior to the previous best one.