I'm experimenting with Open CV, and it seems to come dangerously close to doing what I need right out of the box! Any suggestions as to how to best find and track the retina of the eyes themselves?

not sure if this is exactlyy what you need but did check out :

Processing Easy Eye Tracker Plugin (PEEP)

http://processing.org/discourse/yabb2/YaBB.pl?num=1266256717

Yeah! Looks like a great place to start. Too bad it's PC-only :(

I've got a PC, so I can go that route if it's the best way! Ever used PEEP before? Does it work well?

no never tested it. its pretty new and i never had a good idea/project to test it, but tell us about your experience...

I'm not sure that PEEP is designed to work with a webcam.  From their FAQ:

"...for a truly realistic feeling Text 2.0 can offer you need an eye tracking device, which only very few people do possess right now."

From the downloads page:

"And even if you don't have a tracker yet, PEEP comes with a simulator to help you out. "

TBH I'd be astonished if, with current technology, you could track eye movement/gaze direction with a single webcam:  if you think about it it's an incredibly complex problem.  Interpreting the 3D direction of an object with a 2D camera can be done, there are plenty of 'AR' demos out there; but that usually relies on large and obvious registration points (e.g. graphic blocks on a card).  To do the same with an eye I think you'd need a camera focused in specifically on the eye, so you can pick out all the details and track its movement.  You'd also need to know the exact location of the eye in 3D space in relation to the surface being read: any slight error and your gaze measurement would be way off.  The further the eye is from the surface being read, the larger the margin of error...

I remember trying one of those eye tracking systems a while ago and it relied on the user putting their head in a very specific position, close to the reading surface, presumably so the eyes would be in the field of view of the tracking camera(s).  I think there's a reason why this sort of development is happening in University labs...

Best of luck, but I'd suggest you try for something less ambitious, perhaps just using head position?

Thanks for the discussions, folks. I'm definitely thinking that this will only work well with a) a large "poster", b) a user standing in a relatively predictable close-proximity location, and c) several cameras.

Wikipedia has some great information:

http://en.wikipedia.org/wiki/Eye_tracking

It's not totally out of the question to have the user peer into a little light-box to view the image, thus simplifying the technical hurdle immensely. While this approach would be cheap, simple, and accurate, it would be much less elegant.

One option I haven't yet fully explored is to use processing to manually track the eyes; i.e. scrub through multi-camera video footage by hand and drag a "gaze" locator around to approximate where the user is looking at a given moment. It would be incredibly tedious, but pretty fool-proof (and a breeze to set up!)

Still looking at options.

The wikipedia article is indeed interesting...  Of course I was looking at the problem from my sight-biased perspective and not thinking in terms of other measurable light ranges.  If you can cause the red-eye effect using a non-visible light range you've got a much easier point to track; though even then I think the distance involved with a 'poster on a wall' is likely to be problematic.  It's the classic 'a ship travelling a thousand miles with a tiny error in its bearing will land up way off course' issue: i.e. the further away the user is, the less accurate the results are likely to be.  And you're still going to need either a very high resolution camera; or 'know' where the user's head is likely to be and focus on that area; or have a mobile camera that can track the user's head ( but then of course you'll have to factor in the direction it's facing into the calculations)...  All good fun!

It's certainly an interesting project.  Best of luck with it!

Why don't you use the technology of biometric time clocks such as the eye recognition system.