Most mobile eye trackers assume that the user is always looking at the same plane. This assumption is acceptable when the user’s gaze is limited to a computer screen. However it introduces an error when the eye tracker is used to collect gaze data over 3D objects in natural scenes. A significant error component is due to the parallax between the eye and the scene camera of the mobile eye tracker. In this work we present a new method based on invariant properties of projective geometry to compensate the parallax error using two calibration planes. Simulation results indicate that the method is sound and promising, being more robust to depth variations than traditional methods, with average errors up to 2.5 times smaller.
This work is in pt-BR.