Eye tracking: Difference between revisions
Jump to navigation
Jump to search
mNo edit summary |
mNo edit summary |
||
| Line 4: | Line 4: | ||
* illuminates the inside of the eye with near-infrared | * illuminates the inside of the eye with near-infrared | ||
* IR camera that | * IR camera that | ||
"Can you do this with a webcam?" | |||
At all? Yes. | |||
Accurately? No. | |||
You can download an app for your phone right now. And its output will be... not great. | |||
Why? Multiple reasons | |||
'''You need a comparison.''' | |||
Hold your head still and look around. | |||
Your pupils are now a good indicator of proportional movement. | |||
(You still want to calibrate what the extents are, but that is something you needed ''anyway'') | |||
Now rotate your head and keep looking at the same spot. | |||
Your pupils are moving from left to right, the place it should detect does not move at all. | |||
From ''just'' the image of the pupil in the eye, you would not get a halfway decent direction | |||
unless we clamp some heads in place. | |||
"Can't you add, I dunno, 3D head tracking?" | |||
Yes, and that does help, but also adds assumptions to the whole, as well as another limited-accuracy thing . | |||
The corneal reflection style eye tracking adds a reference that is also free to move - the back of the eye. | |||
This definitely still has limitations, but is | |||
'''Just plain resolution''' | |||
Also, most webcams are wide angle. That leaves a dozen pixels for the pupil. | |||
You can imagine that tracking across thousands of pixels on a screen, | |||
using a few dozen pixels of observable movement, is not going to go well. | |||
Higher res doesn't help a lot due to the optics and typical lighting | |||
and the way various webcams sort of invent a few pixels. | |||
You need either something close to your eyes, or zoomed in. | |||