Rotated bivariate Gaussian model for gaze-based pointing endpoint distribution modeling

Abstract

Endpoint distribution modeling constitutes an important research topic in the field of human-computer interaction, as it reflects the user's psychological state and decision-making process during interaction, and holds significant implications for the design of interaction assistance. We investigate how the direction of gaze movement affects the fixation point distribution and how this factor can be incorporated into the modeling process. We propose a rotated bivariate Gaussian model, which represents an improvement and generalization of the bivariate Gaussian model, enabling the incorporation of gaze movement direction into fixation point distribution prediction. In this model, the major axis of the prediction ellipse, such as the 95% prediction ellipse, is aligned parallel to the gaze movement direction, while the minor axis is perpendicular to the gaze movement direction. Additionally, projected target width and height are used in place of nominal target dimensions for fixation point distribution modeling. Compared with traditional distribution models, the proposed model better captures the observed phenomenon of fixation point distribution elongation along the gaze movement direction, thereby enhancing the accuracy of eye-tracking-based interaction.