Email updates

Keep up to date with the latest news and content from JNER and BioMed Central.

Open Access Methodology

A free geometry model-independent neural eye-gaze tracking system

Massimo Gneo*, Maurizio Schmid, Silvia Conforto and Tommaso D’Alessio

Author Affiliations

Engineering Department, “Roma Tre” University, Via della Vasca Navale, 84, Rome, I-00146, Italy

For all author emails, please log on.

Journal of NeuroEngineering and Rehabilitation 2012, 9:82  doi:10.1186/1743-0003-9-82

Published: 16 November 2012

Abstract

Background

Eye Gaze Tracking Systems (EGTSs) estimate the Point Of Gaze (POG) of a user. In diagnostic applications EGTSs are used to study oculomotor characteristics and abnormalities, whereas in interactive applications EGTSs are proposed as input devices for human computer interfaces (HCI), e.g. to move a cursor on the screen when mouse control is not possible, such as in the case of assistive devices for people suffering from locked-in syndrome. If the user’s head remains still and the cornea rotates around its fixed centre, the pupil follows the eye in the images captured from one or more cameras, whereas the outer corneal reflection generated by an IR light source, i.e. glint, can be assumed as a fixed reference point. According to the so-called pupil centre corneal reflection method (PCCR), the POG can be thus estimated from the pupil-glint vector.

Methods

A new model-independent EGTS based on the PCCR is proposed. The mapping function based on artificial neural networks allows to avoid any specific model assumption and approximation either for the user’s eye physiology or for the system initial setup admitting a free geometry positioning for the user and the system components. The robustness of the proposed EGTS is proven by assessing its accuracy when tested on real data coming from: i) different healthy users; ii) different geometric settings of the camera and the light sources; iii) different protocols based on the observation of points on a calibration grid and halfway points of a test grid.

Results

The achieved accuracy is approximately 0.49°, 0.41°, and 0.62° for respectively the horizontal, vertical and radial error of the POG.

Conclusions

The results prove the validity of the proposed approach as the proposed system performs better than EGTSs designed for HCI which, even if equipped with superior hardware, show accuracy values in the range 0.6°-1°.

Keywords:
Eye-gaze tracking; Human computer interaction; Pupil center corneal reflection; Artificial neural networks