Low-cost Remote Eyetracking

[andronickvll]

My brain is full of fuck from the youtube comments...

erm? boobs? guys...?

[piechulla1966]

@andronickvll

???

[NVaizard]

Anyone knows if there is an ideal type of webcam for eyetracking? Maybe the ones with better frame rate, or is it only software related?

[piechulla1966]

@NVaizard

Maybe the PlayStation Eye?

[NVaizard]

@piechulla1966 Yeah, I tought about it too. But I was wandering if there was a "technical why", like when you buy a TV you go for the one with greater resolution, but when you go for a webcam looking for eye tracking what should you look for?

It bugs me...

[piechulla1966]

@NVaizard

1. Focal length: most webcam's optics have a wide angle characteristic. You do not want this for eyetracking. With the PS3 Eye you switch between two focal lengths, one of them is interesting for eyetracking. 2. Sensor quality: Hard to measure, but the PS3 Eye has a contrast that seems not so bad at all. 3. Speed: There is a rumor that says you can get 640*480 @ 60 fps with the PS3 Eye -- did not work for me, maybe somehow i fucked up the third party Windows PC driver installation.

[piechulla1966]

@NVaizard

4. Option to enable near infrared (NIR). If you manage to remove the IR blocking filter, you get a pitch black pupil, which is easy to track. This manipulation is impossible with most webcams, very easy with some webcams (e.g. MS VX-1000 and MS VX-6000), and hard with the PS3 Eye (I didn't even try it -- but others did).

[piechulla1966]

@NVaizard

Reality check: Why using a webcam at all? I am playing with a Thorlabs DCC1545M camera with Tamron 12VM412ASIR optics these days. No need to remove the IR blocking filter: Professional B/W cams do not have an IR blocking filter. Why breaking an expensive webcam, if can buy what you really need if you invest just some additional bucks?

[NVaizard]

@piechulla1966 Thanks for the answers, I'm searching for different applications of new technologies for education. Eye tracking and head tracking are strong candidates. I'm starting to give up on eye tracking simply because it looks too underdeveloped and hard to work with at this moment. But thanks a lot for your answers, they were very helpful :)

[piechulla1966]

@NVaizard

Oops! You are a consumer and/or researcher, not a developer. Well, there _are_ off the shelf products. You could test them. But they are expensive, because this is a niche market. On the other hand, there are also Open Source projects (e.g. the ITU Eyetracker). If you are working on an academic project, you can try to cooperate with this research group. But eye tracking for education? A blackboard and a piece of chalk should be much more effective (IMHO).

[NVaizard]

@piechulla1966 lol you kind of have a point there. :) But the reason it would work is the novelty factor. Kids in particular respond very well to new technologies, so when we apply that to educational purposes it comes back with very good results :). Anyway, thanks again for the suggestions, but we will probably look for a less sofisticated fom of tracking, like head, hand, color and shape tracking, maybe allied with gesture recognition. Any tips would be very welcome, see ya :)

[piechulla1966]

@NVaizard

So, Kinect (formerly project Natale) should be very interesting. It seams to be a genuine TOF ("time of flight") cam (cool stuff). This device was hacked immediately. So, both Prime Sense and Microsoft changed their minds and published a programming API. Kinect is a big success in gaming these days. I feel that nobody knows if it can also be used for "serious applications". When I saw the movie "Minority Report" I thought "OMG, why doesn't he simply use a mouse?"

[jerrygdm]

Is it so difficult to implement this? I don't know OpenCV, I'm able to program in java and c++. In how much time you learn OpenCV and implement this? Thanks

[piechulla1966]

@jerrygdm

For starters: The knowledge needed to do this has absolutely nothing to do with OpenCV. OpenCV is just a computer vision library. What's nice in OpenCV is that it is free and it has an active community. But I do not use OpenCV much at the moment because HALCON ia much more important for me these days.

If you are interested in machine vision -- A really great text book is:

Bradski, G. & Kaehler, A. (2008). Learning OpenCV. Computer Vision with the OpenCV Library. O'Reilly

[wuhlei]

nice!

[wounsel]

check out Fffffat labs stuff on eyetracking... might be useful

[piechulla1966]

Not bad indeed. Seems to be derived from the openEyes project.

[alfiecarter111]

Have You checked it for different orientations or angles please ? I see only the frontal face and eye detection inside frontal face... please reply

[piechulla1966]

Rotation is not a problem if you use the classic cornea reflex approax. Please google for "cornea reflex". But: This was my first attempt. I was sueprised that I could measure ANYTHING AT ALL :)

[lsd]

@piechulla1966 I googled for "cornea reflex" (google suggested "corneal reflex" instead, so I googled both) and only came up with the blink response--e.g. if you stick a hair in someone's eye, their reflex is to blink; how fast this response is can be a measure of the animal's sedation. Googling for "cornea reflex approx" yielded more relevant results.

[piechulla1966]

@lsd Try "eyetracking corneal reflex" -- overloading is not restricted to object orient programming, you can also find it in natural language!

[piechulla1966]

To Stephenlee5: Sorry for accidentally destroying your comment. To answer your question: Yes, after calibration finishes, it indeed tests for eternity. It's not a bug, it's a feature ;-)

[piechulla1966]

Hi YouTubers, please don't use the "Send Message" feature to ask me questions. It's too buggy (internal server errors most of the time). Instead, please use the email address in the documentation. Thanks!

[piechulla1966]

The download is on-line (ftr_by_wlp_08.zip). See documentation in "About This Video". Merry X-mas and a happy new year!

[piechulla1966]

On-line documentation now available. See "About This Video". Merry X-mas!

[naturalui]

could you explain this a bit more? what software are you running etc..? any documentation?

[piechulla1966]

I wrote it in C++ in August 2007. It's a tiny "just for fun" project -- not worth being documented. BTW, your pupil center detector seems to work pretty well, congratulations!