Eye-project proof of concept

Eye-tracking using visual spectrum cameras and partial-reflection method: Proof of Concept

765 BME C90-0 BME Design

J. Browning, Chang, Morshedzadeh, and Huo

January 11th, 1997

One eye, closeup. Pupil might be visible with image enhancement.

A friend of mine worked in a department lab that had these connectrix quick-cam cameras available. He invited me over to use them to test our idea for a headset, and as soon as he called I went. I'm sorry I didn't get the chance to let you guys know about it in advance; I didn't know myself the camera was available until my friend called me from the lab. Follows are is a summary of that night's work. Please tell me what you think!

The prototype

A sideways drawing of the unit.

The user's eyes would be on the right, the image you're looking at on the left, camera from above.

Two views of the unit. The plasic part is a cd case. The camera lens goes into the square hole. You view from this side (side with the orange pin).

From above. Camera looks into square hole.

Me using the unit. The white thing is the camera itself.

Unit without camera with user wearing it. Camera would go where the arrow is.

Some sample pictures taken by the camera

The quality of these images was lost some during conversion between formats. Also, I was using the inside of the cardboard box as a background. If we mounted a black background (as marked in position (B) on the drawing of the prototype design) or had a more sensitive camera, we might have gotten better results. All of these images are stright from the camera, without any kind of enhancement or touching up.

Picture with camera in camera port. Light from above. Both eyes are visible.

Eye in the upper left, staring straight ahead, and lower right hand corner. This seems to suggest even this crappy design can tell the differences between different quadrants. Again, these images are straight from the camera, without any kind of touching up.

I also shot a movie of my eye moving around, which you can get as a 16 Mb .avi (video for windows) file (4 Mb zipped). E-mail me if you are interested.

The camera itself

Interior plan of the camera. Front is the Adjustible focus lens. Behind that is the Image capture board with CCD, and behind that is the VIDEC compression board. The camera just feeds into the keyboard plug: no other boards are necessary. Camera is self-contained; all of the electronics needed to make images, capture the images, and translate them to the computer is on board (I think. It may only be compatible with certain video cards; that's one thing we need to check.).

This is the picture of what the camera looks like. We could shave off the outer frame.

Demonstration of optical viewing quality

Your eyes would be where the blue thing is. The object being viewed through the headset is the pyramidal lump at the right.

What the person sees. Note that the CD case does not appear to interfere siginificantly with the quality of the image being viewed. This image is the one that follows the purple arrow in the drawing of the prototype design.

What the camera sees via the reflecion off the CD case on the left, what it should look like on the right. The above image is the image formed along the red arrows (see below drawing again.)

What's next?

Lots of things..here's what I've thought of so far, if this really does work.

The image quality needs improving. Better cameras and a better background would help.
We need a better prototype (I had to give back the CD case and push pins, so I don't have the prototype anymore.)
Also, finding the actual components for the head-mounted display, the wearable computer, etc. may now be in order.
We need to think about what software exists to actually do the calculations.
What level of accuracy is acievable is unknown. I think this incredibly primative design can do quadrants (certainly left/center/right) but we can't yet resolve a pupil yet.

Something that is really bothering me is why noone else has done this type of design before. None of the patents (through 1976) whose abstracts I looked at do eye-tracking this way (I think: the patent abstracts were really obtuse reading! :) ) It may be because cameras this small and computers this powerful were not available until a year ago. Also, the idea is so simple that it would not be hard to circumvent any patent for a design of this type (it would be easy to add enough refinements to make a new design different enough from the old to get a new patent.) There has to be something we've missed as to why this simple design can't work...if you guys can find it, or have new ideas, please don't hesitate to share! I'm sorry again I couldn't reach you guys before I went to lab: if I could have known I would have had the chance to do these experiments, I would have let you guys known. See you in class!