Estimating direction and depth of visual fixation using electrooculography

Eye tracking is a useful method of human-computer interaction and of interest to scientific studies on cognition. An electrooculogram (EOG), measuring the changes caused by the angular rotation of the dipole moment of the eyes, contains information on the direction of each eye´s movement. A known target stimulus moving in 3D space was used to calibrate an 8-channel EOG to resolve the target position and fixation depth. Linear estimations of target position, valid for small-angle deflections from center, resulted in average azimuth (horizontal) and elevation (vertical) angle errors of 2.9±2.1 and 4.4±3.3 degrees, respectively, and average fixation distance errors of 13.4±11.8%(7.5±5.6 cm). Accounting for nonlinearities from wider angles and closer fixations, estimations by a single layer, feedforward neural network, resulted in average errors for azimuth and elevation angle of 2.3±1.6 and 3.2±2.6 degrees, and average fixation distance error of 10.3±10.0% (5.7±4.7 cm). This work is the first apparent demonstration of EOG signals as a basis for eye tracking in 3D.