Eye-Movement Tracking Using Compressed Video Images

Abstract

Infrared video cameras offer a simple noninvasive way to measure the position of the eyes using relatively inexpensive equipment. Several commercial systems are available which use special hardware to localize features in the image in real time; typically the pupil and the first Purkinje image (corneal reflex) are tracked, the difference between these two signals giving a measure of eye rotation which is relatively independent of head position. While these systems have the advantage of providing a result in real time, this speed is obtained at the cost of reduced resolution and accuracy. The limitation is not imposed by the information content of the video image, however, but rather by the complexity of the processing algorithms which can be implemented in current hardware. More accurate results can be obtained when the imagery is analyzed off-line using more complex algorithms implemented in software.