Video-based approach for tracking eye orientation in three dimensions.
Item
-
Title
-
Video-based approach for tracking eye orientation in three dimensions.
-
Identifier
-
AAI9732992
-
identifier
-
9732992
-
Creator
-
Zhu, Danjie.
-
Contributor
-
Adviser: Theodore Raphan
-
Date
-
1997
-
Language
-
English
-
Publisher
-
City University of New York.
-
Subject
-
Computer Science | Psychology, Physiological
-
Abstract
-
Accurate eye movement recording and monitoring has been growing in importance in the differential diagnosis of diseases associated with vertigo and other neurological diseases. Tracking of eye movements in three dimensions is important for modelling and studying saccadic eye movements and those generated by the vestibular system. The scleral coil system is currently a standard technique for measuring three dimensional eye movements in both animals and humans. However this technique is invasive. To bypass the invasive nature of the coil system in measuring eye movements in three dimensions, there have been a number of investigations on the use of cameras and video-based techniques. The methodology developed by Moore et al. (1996) uses a three dimensional rotation model to find the eye orientations in three dimensions. It is based on finding the orientation of the optic axis by computing the center of mass of the pupil and then computing the rotation about the optic axis utilizing variation in intensity along an iral arc. This methodology is dependent on maintaining a clear and small pupil image so that the correct center position can be found. It is also dependent on maintaining a fixed pupil size in order not to have changes in the iral striation patterns which would cause errors in computing torsional eye position. The purpose of this dissertation is to develop algorithms based on a more global model of how circular contours which inscribe the pupil and iris are mapped to elliptical contours under a rotation and orthographic projection mapping, i.e., affine transformations. We have implemented an edge detection algorithm that determines the contour at each given eye orientation. The orientation of the optic axis relative to the axis of the image plane is computed by finding the best fitting ellipse for this contour. The shape of the ellipse can then be related to the orientation of the optic axis. Torsion about the optic axis is computed by using the a minimum distance pattern matching algorithm on an annulus of the iris. The algorithm allows for the removal of the image noises due to the infra-red reflection from the annulus. In order to obtain a clear and uncovered iral signature, an adaptive algorithm has been implemented to choose the location of the annulus depending on the orientation of the eye. This approach to computing the torsional component of eye orientation tends to be more immune to artifacts which arise when the eye moves into positions where there is partial lid occlusion. The proposed methodology should be important for developing a robust system for tracking eye movements.
-
Type
-
dissertation
-
Source
-
PQT Legacy CUNY.xlsx
-
degree
-
Ph.D.
