Eye Tracking Using Fiber Optics and Computer Vision Techniques

Abstract

We describe a system for tracking the motion and determining the velocity of the human eye using a combination of fiber optics and imaging optics integrated with a CCD camera, a framegrabber, and a PC. Systems such as the one we describe are needed for monitoring location of interest for better human-computer interfaces and virtual reality simulations, for monitoring pilot cognitive skills during flight simulations, and for detecting the early warning signs of driver drowsiness. Where some eye-tracking devices are cumbersome because they require a person to limit head motion to remain in view of a camera or uncomfortable because they require the use of an elaborate headpiece, our proposed system is as simple to wear as a pair of eyeglasses, with the only added weight to the glasses being that of a fiber optic bundle.

This fiber bundle guides the image of an 880 nm LED illuminated eye as it is reflected from a hot mirror built into the eyeglass assembly. The guided image is focused onto a CCD and the camera’s output is linked to a PC framegrabber card. Software written in Microsoft Visual C++ locates the user’s pupil in each video image and tracks the motion and velocity of the pupil over time.

The prototype system tracks the center of the pupil to an accuracy of only 1/5 of the pupil size. This diminished accuracy results from the minimized computation scheme of pupil detection. The technique relies on edge detection and direction of image gradient over very few pixels so that the system is optimized for speed. Many other methods of pupil detection could be investigated in future research in order to optimize such a system for other parameters.