Eye Tracking in Videos (Research) Question

[ananth]

I am a researcher doing an eye-tracker experiment. The apparatus consists of a 640x480 avi file that is displayed on a 1024x768 (not sure about the monitor resolution) monitor. The video consists of several 10-second clips of humans or dogs or objects. We present these videos in sequence to our human subjects, and we track where the subjects are looking at using an eye tracker.

The eye tracker is a camera that tracks where the eye is looking at on the screen through periodic calibration, and keeps track of it. It outputs 2-dimensional coordinates that reveal exactly where the subject's eye was looking at on the screen at 60Hertz.

My task is to be able to track the location of each eye on the 10-second clips. This is made particularly difficult because these eyes move within the 10-seconds. Additionally, I need to come close to matching the 60Hz rate at which the eye-tracker logs the information. I am at a loss. How can I start? Thanks.
- Ananth

[Guest]

Do you mean you need to translate the coordinate data to a marked track on the video?

One could without much difficulty write an Avisynth filter to do that. You can interpolate the tracked points to resample for the AVI frame rate. Then just stick a blob on the video at the interpolated points.

If you post a link to a file and corresponding tracking points, maybe I can whip it up for you.

I didn't understand this part:

"This is made particularly difficult because these eyes move within the 10-seconds."

Isn't the whole point of the experiment that the eyes move? If they didn't move why would you need to track them?

[ananth]

I didn't understand this part:

"This is made particularly difficult because these eyes move within the 10-seconds."

Isn't the whole point of the experiment that the eyes move? If they didn't move why would you need to track them?Regarding your last question, I guess I was being overly redundant. What I meant to say was that the eyes of the dogs are particularly prone to motion.

Do you mean you need to translate the coordinate data to a marked track on the video?

One could without much difficulty write an Avisynth filter to do that. You can interpolate the tracked points to resample for the AVI frame rate. Then just stick a blob on the video at the interpolated points.

If you post a link to a file and corresponding tracking points, maybe I can whip it up for you.That is not what I want to do. One set of coordinates is the coordinates that the subject looks at, that comes from the eye tracker. The second set of coordinates are where the eyes of the human or dog are located on the video, or in some cases, the object. We need to compare the two coordinates to get a statistical measurement of whether the subject is looking at the eyes, and compare it across different subject populations and conditions.

The first set of coordinates, of where the subject is looking at, I can easily extract from the data I have.

For the 2nd set of coordinates, I want to be able to extract coordinates (of physical location of eyes or objects) from the video files. This is what I am having a considerable amount of difficulty with.

The video files are 640x480 and the monitor resolution is different, so I have to account for that as well.

Hope this explanation helps. I can still put a link to one of the videos if that'll help. Thanks for your quick reply.

[Guest]

That's much more difficult. :(

You need a coordinate position digitizing system:

http://www.google.com/search?sourceid=navclient&ie=UTF-8&rls=GGLD,GGLD:2005-08,GGLD:en&q=coordinate+position+digitizing+software

[ananth]

That's much more difficult. :(

You need a coordinate position digitizing system:

http://www.google.com/search?sourceid=navclient&ie=UTF-8&rls=GGLD,GGLD:2005-08,GGLD:en&q=coordinate+position+digitizing+software

I tried the link and several companies and several products. None of the actually seem to have products that digitize video files. I finally sent an email to silicon graphics.

This actually got me to thinking. My understanding is that you only need to digitize images if delineation was needed. I am wondering if I can't do this manually. Is there a way that I can manually select the regions and use a program to extract the coordinates? Thanks.
- Ananth

[Guest]

It would be possible to make a simple GUI that lets you step through the frames one by one and record mouse click positions as you click on things. That's what I meant by coordinate digitizing.