I worked with eye tracking software a decade ago and one of the most interesting and tricky problems was your eye would look at the cursor and over compensate, causing the cursor to keep drifting. It was in essence user error but oddly difficult to control. Does anyone know if this has been solved for?
I worked in a team that researched how computor vision might intersect with art and aesthetics. We worked with eye trackers to see if artists look at images any differently to non-artists. The difference was in the way that artists attention would frequently be caught by incidental background items (highlights, corners, features of varying kind). Non-artists would dwell on the immediate content... in the case of people, mostly faces.
I’m thinking of ways that project can have application the this one. Perhaps to train people to look in different ways? Certainly it has the potential to open up the process of drawing to matriculation.
Several other comments are alluding to saccade [1], where your eyes unconsciously jerk around (saccade is the french word for jerk) any vision control system would have to allow for this.
In the simplest approach this would be a low pass filter, presumably you could throw a bunch of ML to filter them out more accurately
Saccades are under voluntary control. They're the fast movements when you switch where you look.
There are microsaccades that are similar but don't seem to have any target we understand. In general these, and other microeyemovements like tremors and drifts, are not very well understood, and even their existence can be somewhat quoestioned.
Low pass filters are somewhat problematic for eye movements because the altenating slow and fast movements.
> Saccadic eye movement allows the mind to read quickly, but it comes with its disadvantages. It can cause the mind to skip over words because it doesn't see them as important to the sentence, and the mind completely leaves it from the sentence or it replaces it with the wrong word. This can be seen in 'Paris in the the Spring'. This is a common psychological test, where the mind will often skip the second 'the', especially when there is a line break in between the two.
Well, looks like I just failed the test, even though I made a genuine effort to read the paragraph very carefully after the the second sentence.
Impressive what can now be done with such precision with just a regular webcam. Previously special hardware like the Tobii eyetracker was required for this level of precision.
eyes naturally jump and lock onto things. They don't just slide around between focal points. So... probably neither a problem with the artist or the program.
It's a hilarious thing to say: the fact is eyes are just really bad at drawing and not evolved for that purpose