Open AccessSelection of unique gaze direction based on pupil position
Author(s) -
Mohammadi Mohammad Reza,
Raie Abolghasem
Publication year - 2013
Publication title -
iet computer vision
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.38
H-Index - 37
eISSN - 1751-9640
pISSN - 1751-9632
DOI - 10.1049/iet-cvi.2012.0141
Subject(s) - ellipse , computer vision , artificial intelligence , computer science , gaze , iris (biosensor) , position (finance) , eye tracking , iris recognition , boundary (topology) , selection (genetic algorithm) , pupil , manifold (fluid mechanics) , mathematics , geometry , optics , mechanical engineering , mathematical analysis , physics , finance , biometrics , economics , engineering
The ‘gaze estimation’ problem, because of its manifold applications including human–computer interaction especially for the handicapped, has been a topic of research for many years. Recently, thanks to technological advances, non‐intrusive methods based on image processing employed in broader applications, are addressed more than before. One of the promising approaches to gaze estimation is based on projective geometry. In projective geometry‐based approaches, an ellipse is fitted to the image of iris boundary and from its parameters, two solutions are obtained, only one of which is valid for the gaze direction. Since the ellipse parameters are not adequate for disambiguation, in a previous work, the accurate coordinates of eye corners in three‐dimensional, as complementary information, is obtained through another system increasing the cost and complexity of the overall system. In this article, a new technique to select the valid solution, based on eye geometry and iris image, is developed. In the proposed technique, relative position of the pupil centre with respect to the iris centre is used as the complementary information and a novel algorithm is proposed for its extraction. The performance of the proposed technique was evaluated on 600 real images and with only one failure on selecting the valid solution demonstrated an accuracy of 99.8% for disambiguation.