Robust and real-time torsional eye position calculation using a template-matching technique

Computation of eye rotation about the line of sight (torsion) using image processing techniques has traditionally used cross-correlation of iral signatures sampled from circular arcs centered on the pupil. We have developed a new algorithm that utilizes a template-matching technique to calculate torsional eye position. Iral signatures are obtained from two annuli centered on the pupil center. By assuming that torsional rotation of the eye is constrained between successive video frames (<2°), only a small window of the previous reference signature is necessary to determine relative torsional eye displacement. This dramatically reduces the number of pixels needed for computing torsion. This algorithm is considerably faster, attains a higher accuracy, and exhibits considerably less noise than the cross-correlation technique. Running on a 800 MHz Intel-based Dual Processor Pentium III, with a Matrox frame grabber, the system is capable of processing three-dimensional eye position at a rate of 120 frames/s.