Input Evaluation of an Eye-Gaze-Guided Interface: Kalman Filter vs Velocity Threshold Eye Movement Identification

Abstract

This thesis evaluates the input performance capabilities of Velocity Threshold (IVT) and Kalman Filter (I-KF) eye movement detection models when employed for eyegaze-guided interface control. I-VT is a common eye movement identification model employed by the eye tracking community, but it is neither robust nor capable of handling high levels of noise present in the eye position data. Previous research implies that use of a Kalman filter reduces the noise in the eye movement signal and predicts the signal during brief eye movement failures, but the actual performance of I-K.F was never evaluated. We evaluated the performance of I-VT and I-KF models using guidelines for ISO 9241 Part 9 slandard, which is designed for evaluation of non keyboard/mouse input devices with emphasis on performance, comfort, and effort. Two applications were implemented for the experiment: 1) an accuracy test 2) a photo viewing application specifically designed for eye-gaze-guided control. Twenty-one subjects participated in the evaluation of both models completing a series of tasks. The results indicates that IKF allowed participants to complete more tasks with shorter completion time while providing higher general comfort, accuracy and operation speeds with an easier target selection than the I-VT model. We feel that these results are especially important to the engineers of new assistive technologies and interfaces that employ eye-tracking technology in their design.