Tobii and SteelSeries Release Consumer Eye Tracker

After both my supervisor and I took a good long crack at it, we could think of very few ways it could be used. Of course, eye tracking has traditionally been used in the medical disabled industry, for good reason.

But when it comes to "What can it be used for in a university setting", we came up with a blank.

rdw is correct about the eyeX, and it also applies to the Rex: the operational area is around 2 foot cube. If you move your eyes out of that, the driver/SDK throws a tantrum and desyncs your gaze. This also means having an interactive demo is nigh impossible, as the gaze tracker has to be within a small area of the head/eyes.

And you better not blink, as it cannot track an eye. It has to have both eyes available. Im not entirely sure why they didn't implement tracking with each eye independently, but that's what they did.

Calibration is also a pain. You have to calibrate the device stick to the monitor (not hard), and then you have to follow your gaze around the screen to the little dots for calibration. With this calibration, it changed significantly from morning to evening to the point I had to be a different user.

I also found their SDK and accompanying demo software to be completely lackluster. Every demo required keyboard button presses. Not only that, but their program in the control panel had no hands-free options; just mostly hands free. There was no completely hands-free app to control parts or the whole computer. This is one of the reasons we thought this might be interesting.

The other issue is of accuracy: what the hardware is doing is sending out tons of 'gaze detection events' that raytrace where the device thinks you're looking. When you're in the calibration, you can see this by the dozens of dots around the 'focus dot'. These are the ray-traces that attempt to figure out your gaze. And it's not great. Some dots in the middle (think + on the screen) were fine. But the dots on the 4 corners of the screen would routinely come in as scattered clouds that would "kind of, not really" synchronize. And we'd have to go back through the calibration because accuracy really stunk.

In the end, our dept gave our Tobii Rex to the Disability office of the Chancellor. We both hoped they could tease out better performance and usability than we could.

The main limitations that keep me from using it:

- Mac is not supported.

- Very small active area. You basically have to keep your head within about a 2x2 foot cube. You can't lean back, on an armrest, or even slump.

I've had the chance to talk to a few Tobii-engineers in Stockholm at a few general gatherings and like I previously said, Tobii seems to be a pretty interesting company.

I've heard that Tobii Technologies is going to launch an Initial Public Offering on the NASDAQ OMX Stockholm Stock Exchange sometime under 2015, although official time plans are not available.

Investor AB owns about 20% and Intel owns about 10% of Tobii Technologies.

1. Performance! You only need to render high-quality pixels in the places where the user is (or will) look. Lower-quality content can be put into their peripheral vision and they won't notice.

2. Depth-of-field effects will actually be good instead of a pain in the ass, because the "focal distance" is more likely to be whatever the user is actually looking at. (Not perfect, though: Is the user looking at the semitransparent steam-puffs 5m away or the wall 100m behind?)

3. High dynamic range can behave and adjust appropriately since the system knows whether you're gazing into a light or dark area. Current systems either take the entire scene into account (bad) or weigh things towards the crosshair (a bit better) which might not be where the user is actually trying to look.