Error Reduction in 3D Gaze Point Estimation for Advanced Medical Annotations

The advanced surgical navigation and computerbased anatomical education systems require the detection of gaze point of a user to detect the intention of the user. When much information is presented using AR annotation, a user may not obtain the information efficiently. The displaying of selected annotations increases the visibility of the annotations by reducing the amount of information. However, when viewing a large area, the depth error of the gaze point estimation becomes much larger than the horizontal and vertical errors. In this study, a method that estimates the gaze point by calibrating the depth with the distance of the Purkinje images was proposed. The result of the experiments showed a significant decrease in estimation error and suggested the possibility of improving visibility with the proposed estimation method in comparison to the traditional one.