• computer vision
• eye tracking
• gaze estimation
• high accuracy
• proton clinic
• proton therapy
• regression
• uveal melanoma
• vision

Uveal Melanoma is the most common intraocular tumor in humans, and one
of the most promising treatments available is proton therapy. Proton clinics
use specic devices for high energy proton beams forming and delivery, used
to damage selectively the tumoral cells, saving the healthy part of the eye
and maintaining as much vision as possible.
Accurate energy dose delivery is critical for this kind of treatment, and
position and rotation of the eye must be measured with high accuracy, in
order to determine the position and orientation of the tumor in real-time.
This is achievable with eye tracking.
Aim of this thesis is the design and validation of a non-invasive eye tracker
suitable for proton clinic environment, with accuracy of 0.5° in the worst case.
For this purpose a feature-based video-oculography eye tracker with IR
active illumination was studied, simulated and implemented. In addition, a
custom pupil and glint detection algorithm has been developed, along with a
Kalman filter. A mapping procedure and a regression strategy has also been
developed and implemented.
Results of a single mapping on 5 volunteers show accuracy above the 0.5°
bound, with great differences among the different tests. Results are below
the required accuracy only if the mapping is repeated several times, and then
averaged.