A robust algorithm for distorted pupil localization based on Ellipse Difference and PSO

Abstract

Pupil localization is the key step of eye tracking, which directly affecting the precision of eye tracking. Through the process of eye tracking, the shape of pupil is usually distorted, leading to the invalidation of traditional algorithm based on circle localization. Moreover, current algorithms of distorted pupil localization are easily interfered by eyelid and faculae. To solve these problems, this paper raised an operator of Ellipse Difference (ED) to avoid the interference of eyelid and faculae. Meanwhile, to overcome the difficulty of ED operator's high order of complexity, this paper associated the algorithm of Particle Swarm Optimization (PSO) with ED to realize fast pupil localization. The experimental results have demonstrated ED-PSO is more robust than current algorithms and can realize real time processing, which has been applied into real eye tracking system.

SET: a pupil detection method using sinusoidal approximation

“…Considering the low Gaze-Tracker's detection rate of 28.48% and 32.75% for Natural and CASIA-Iris image collections, respectively, vs. 85.23% and 83.41% for SET, we argue that this method sacrifices accuracy to a high extent in order to achieve higher speed. SET is still much faster than Starburst and some other methods such as the one proposed by Deng et al ( 2010 ).…”

“…Many low-cost image capture mechanisms have been proposed, mostly composed of off-the-shelf components (Li et al, 2006 ; Schumann et al, 2008 ; Lemahieu and Wyns, 2011 ; Huang et al, 2013 ). These methods either use new approaches (Mäenpää, 2005 ; Asadifard and Shanbezadeh, 2010 ; Lemahieu and Wyns, 2011 ) or use an open-source project such as Starburst and ITU Gaze Tracker (Ryan et al, 2008a ; Villanueva et al, 2009 ; Deng et al, 2010 ; Tafaj et al, 2011 ) (for Starburst and Gaze-Tracker methods refer to Li et al ( 2005 ) and San Agustin et al ( 2010 ), respectively).…”