Towards an EOG-Based Eye Tracker for Computer Control
David W. Patmore, R. Benjamin Knapp · 1998 · Proceedings of the Third International ACM Conference on Assistive Technologies (Assets '98) · doi:10.1145/274497.274533
Summary
This paper describes the development of an eye tracking system for computer control based on bio-electrical signals, intended as a pointing device for people with physical disabilities. The system combines two complementary biosignal approaches: the Electrooculogram (EOG), which measures the DC potential across the eye that varies linearly with eye rotation, and Visual Evoked Potentials (VEPs), which are the brain's electrical response to visual stimuli measured from the occipital cortex via EEG. The EOG provides very fast response (under 0.25 seconds) for tracking eye movement direction but suffers from baseline "drift" — a slowly changing signal offset caused by electrode contact changes, temperature, and other factors — that makes absolute position tracking unreliable. The VEP provides absolute position information through a feedback mechanism: flashing stimuli on screen produce a strong VEP response when the user's gaze is aligned with the stimulus, confirming gaze position. However, VEP is slow (average selection rate of 1.2 seconds) and only works when the eye is fixated. The authors combine both signals in a feedback circuit where EOG provides the fast "feed-forward" cursor movement and VEP provides the calibrating "feed-backward" position confirmation. The system uses gel electrodes on a headband — four EOG channels (vertical and horizontal for each eye) and one VEP channel over the occipital lobe.
Key findings
The authors tested two techniques for handling EOG drift: an adaptive fuzzy logic system using the first and second derivatives of the EOG signal to calculate a stability term (epsilon), and a combined approach adding the VEP-derived position term (v). The fuzzy logic epsilon term proved highly effective at reducing drift — it approaches one when the eyes are fixed and zero when moving, allowing the system to distinguish between genuine eye movement and signal drift. Adding the VEP term provided further improvement, particularly for detecting when the eye was stationary. Three experimental protocols were tested: continuous stimulation during saccadic movements (worked well for both EOG and VEP), stimulation briefly turned off during movements (EOG showed more drift but VEP discrimination improved), and slow smooth pursuit tracking (EOG performed poorly because drift and movement signals became indistinguishable, but VEP phase changes tracked the slow movement effectively). The multiplied combination of epsilon and v outperformed additive combination, producing "stiffer" cursor movement that filtered out small jittering movements around a fixation target without creating unstable feedback loops.
Relevance
This paper documents an important approach in the evolution of eye-tracking assistive technology — using bio-electrical signals rather than camera-based systems. While modern eye trackers predominantly use infrared cameras and computer vision (e.g., Tobii, EyeTech), EOG-based systems remain relevant for specific use cases: they work with closed eyes, are less affected by lighting conditions, don't require a camera in the line of sight, and can be embedded in wearable devices like glasses. The fundamental challenge the paper addresses — signal drift corrupting position estimates — remains the primary obstacle for EOG-based interfaces. The fuzzy logic approach to drift compensation and the concept of combining fast-response movement tracking with slower absolute position confirmation represent signal processing strategies applicable beyond eye tracking to any bio-electrical control interface. For accessibility practitioners, the paper illustrates the trade-offs between different eye tracking technologies: camera-based systems offer better absolute accuracy but require specific hardware positioning, while EOG-based systems offer faster response and hardware flexibility but require drift compensation.
Tags: eye tracking · electrooculogram · alternative input · motor disability · bio-electrical signal · assistive technology · visual evoked potential · fuzzy logic