I'm very near of the final result but i don't have enough time in hands to continue my work straightforward.
I also had to abandon the idea of getting pupil movement in relative position to the head, this is damn too complex to handle and would require much more than a M10000 to calculate.
I adopted another way of filtering which is even better than contours estimation. When the tracking is locked it remains locked regardless of the eye speed and direction.
Apart of this i have the code to make the bridge between eye moves and cursor moves, need a little time to make the junction with those two codes and that's it
So here is the video, the first minute is garbage as my camera was not aligned to the screen, sorry
As it could be difficult to understand what happens in the video, you will notice the pupil is very dark (black), in it's original state the pupil density is not so high and the tracker with a threshold level can adjust this density. A fine adjust can make the tracker to not take the background, we also do not rely on the infra-red reflections which would always be out of the center of pupil so they would make false data.