Eye-Position Data not accurate

[Eike Froehlich]

Hi,

i am working on a project which uses the eye-position to control an object (… what the eyes do control does not matter in this case).
To get the eye-position data, i use Tobii EyeTracker 4c, Core SDK and the “EyePositionStream” (C#) .
In this application i am using the EyeTracker without any kind of screen.

I’ve mounted the EyeTracker in front of the user, right where you would put it if you were using it for gaze tracking on screen.

The stream works fine. I can see the eye-position data. The data changes as expected when the user moves his head.

But after a few tests i found out that the data isn’t accurate.
I’ve measured the eye-position in relation to the EyeTracker coordinate-system. Comparing the EyePositionStream-data to the measured data, both values differ by more than five centimeters.

Has someone worked with the EyePositionStream without any screen and had this kind of problem too?
Could this problem be caused by no/wrong calibration? (Since I am not using a screen, calibrating is not possible)
Could there be any other cause for inaccurate data?

I would be very grateful if someone could help me.

If you need further information, I will provide it.

[Grant [Tobii]]

Hi @efroe, thanks for your query.

Firstly, it is absolutely crucial that you calibrate the Eye Tracker properly. Not only with an attached screen as intended and designed but for each individual user also.

Changes in eye structure from person to person although subtle, lead to significant changes in the eye tracking algorithm so certainly there is not (or will be) a ‘generic’ profile you can use.

In terms of working without a screen, the Tobii Tech Range of Eye Trackers (4C, Eyex, etc) are designed to work on a screen setup *only* so any change to this setup will inevitably lead to problems of accuracy. It is possible to remove the screen after calibration and replace it with a similarly angled and sized alternative stimulus (say, a book), but again this is not a procedure we can support. For these kind of eye tracking studies, you would need a Tobii Pro range of tracker that does indeed allow for such a custom solution.

Once you have run calibration properly according to the specifications I have outlined, please let us know should your issues persist.

[Jasmine Zoest]

Hi Grant,

I am just going to check that there hasn’t been any updates since this post and that it is still not possible to use Tobii 4C off-screen?

My team has designed a gantry system to move a visual target in 3-dimensional space and would like to use the Tobii 4C to measure eye position in x, y, z. Is the 4C not suitable for this task? We have done some pilot testing and can use a screen to calibrate within the testing area, however, I was hoping to create a calibration routine using the actual stimulus.

Any recommendations would be much appreciated!

[Grant [Tobii]]

Hi @jzoest, thank for getting in touch. The Tobii 4C is part of the Tobii Tech business department which is focused exclusively focused on gaming and interaction and as such we will not support the use of this range of eye trackers in a mobile (non-screen) setup.

Additionally, the use of the 4C for anything other than gaming or interaction purposes requires the purchase of a special licence to do so.

Accordingly, should you wish to perform eye tracking on real-world stimulus, I would refer you to Tobii Pro who specialise in this sector @ https://www.tobiipro.com/product-listing/

Best Wishes

[Author]

Posts