Sclera recognition for eye gaze

[ngonguyen]

Hello all,
I am doing a project that recognize the eye gaze. After I got the eye, I don't know how to analyze the sclera whether a person look on the left or right. I am going to calculate the pixels of the sclera.


So I would like to know that there are any algorithm to solve this problem ?
I am pleased if you help me.
Thank you,
NN.

 

[BradtheRad]

* Do you have an algorithm to determine whether the eye is looking up or down?

* You can count light pixels to assist in reading the gaze, but it may also be necessary to locate the center of the iris. Find the curved edges of the dark area (iris) and calculate the circle's center.

* Set the reference when the eye is looking straight ahead. Calculate the circle's center. Compare all snapshots to this reference.

 

[ngonguyen93]

* Do you have an algorithm to determine whether the eye is looking up or down?

* You can count light pixels to assist in reading the gaze, but it may also be necessary to locate the center of the iris. Find the curved edges of the dark area (iris) and calculate the circle's center.

* Set the reference when the eye is looking straight ahead. Calculate the circle's center. Compare all snapshots to this reference.

Hello,
Thank you for your suggestion.
I have the algorithm for detect the aperture of the eye (looking up and down at this moment), as well as the iris's center. I just get stuck on the horizontal direction. The problem is when I set a threshold for the light pixels, it maybe a mistake (since it also counts the pixels for the region around the eye).
So do you have any idea about finding the threshold (count the light pixels)? I tried using histogram to find but it's not really good.

 

[BradtheRad]

It's an accomplishment just to do what you've done, as far as detecting areas of light pixels vs dark pixels, counting the number of light and dark, etc. Evidently this only gives you partial information. In addition it appears you need to perform other methods. Example, find the extreme left and right corners of the eye, and measure position of the iris between them.

Researchers have made devices that detect which key the eye is gazing at, on a virtual-reality typewriter. The eye movements are no doubt miniscule. The algorithms need to be very sophisticated, and accurate, and use expertly done tricks, to be that accurate.

 

[andre_luis]

There are many approaches that you could follow ( e.g scale/angle variant or invariant ) what expand the possibilities of implementation, but regardless of the choice of detection algorithm itself, you should make a preliminary calibration procedure for each person, in order to take several sample pictures with what you'll map the patterns that will support the training process for each angle.