Robust gaze estimation in low-resolution eye images

A Head Mounted Display (HMD) with a bidirectional chip is an emerging technology with many applications enabling us to display information to users while simultaneously analyzing eye movement [Baumgarten et al. (2012). Aspects of a head-mounted eye-tracker based on a bidirectional OLED microdisplay. Journal of Information Display, 13, pp. 67-71]. However, due to the high density of light emitting and capturing elements, the eye image suffers from noise artefacts and low-resolution output (128\'0296 pixel). Two topics are addressed by this work: low resolution eye tracking and comfortable calibration. Robust eye tracking includes detection of corneal reflections, classification of closed eyes and pupil extraction. Therefore, several different approaches are compared, an extended approach presented and evaluated. Robustness of the proposed approach is increased by incorporating temporal information. A final evaluation shows that implemented extensions improve accuracy significantly. A calibration process is required to estimate gaze points of a user. The second part of this work is about evaluating different ways of calibration methods and presenting novel methods which are more comfortable. The standard multiple point calibration procedure is amended by using automatic methods and animated objects. Finally an implicit calibration by reading text is presented, which makes the disturbing calibration procedure almost invisible to the user with only slight degradiation of accuracy.