Approaching Ground-Truth EntrancePupil Position For Mixed Reality Applications And The Possibilities Of Using Computer Vision To This End

Abstract: Given the requirement to increase the accuracy of eye tracking for mixed reality applications, it was necessary to develop a technology that could more accurately approach the ground-truth position of a human entrance-pupil. The goal was to use Computer Vision, in the forms of pairs of stereo-images of the eyes, to estimate said position of the entrance-pupil. The stereo-vision measurement system proved to be a major obstacle in the process of extracting measurements. This lead to the need for testing the feasibility of stereo-vision for mixed reality, with experiments being performed on an artificial eye. It was established that the geometry of the stereo-system affects how any unit of error will affect the estimations of 3D-positions of any object. A slew of other factors however can lead to errors in the extraction of corresponding points. These errors had their origin mainly in Image Processing and in distortions that were likely caused by the lens of the camera. What this thesis provides is therefore insight into the different sources of error while performing measurements with stereo-images and how these affect the final estimation.