A High Speed Eye Tracking System with Robust Pupil Center Estimation Algorithm

This paper presents a new high-speed head- mounted binocular eye position measurement system using infrared lighting and image processing technology. Current eye tracking systems either run on-line at a lower speed, do the processing off-line, or use dedicated hardware to reach high on-line processing rates. The eye position measurement system we developed only uses a general-purpose computing system and off-the-shelf imaging devices. The binocular system can provide on-line horizontal and vertical measurement at a speed of 150 Hz on a desktop system with a 3 GHz Pentium processor. We report a novel dual-mode software system and a two-step processing algorithm to increase the processing rate. A symmetric mass center algorithm is designed to provide more accurate measurements when the pupil area is partially occluded. Using synthetic eye images, we show that our algorithm performs consistently better than some of the widely used algorithms in industry.