How is precision affected by the calibration version used?

[Michael Bolund]

We have while developing a program using your technology come across a few different versions of software for the eye trackers, and especially in version 1.2 235 OEM we noticed a much shorter calibration process compared to for example version 1.6 641 OEM. We have tried both on our users and haven’t noticed any particular big difference in precision, but as our primary user group is children we prefer the shorter one as it’s easier to keep them concentrated if the process is short. Also the older version (1.2 235 OEM) doesn’t ask for information about correction lenses?

Our question is if there is a precision difference between the the longer and shorter calibration process? Which one is recommended to use(we presume the newest one available is the best one)? Is there any way to shorten the process without a too big loss of precision?

[Anders]

Hi Michael,
that’s a fairly complex question, or set of questions. I’ll address them one by one.

(Note: the version numbers refer to the Tobii Installation and Configuration Tools that shipped with the Tobii REX development kit.)

Q. Is there a precision difference between the the longer and shorter calibration process?
Yes, there is. The more calibration points, the better accuracy you will get — at least in theory. But then there’s the issue with keeping the user’s attention.

A test report from a test performed on nine people in a lab environment shows that the accuracy dropped from 0.41 degrees at 9 calibration points to 0.55 degrees at 5 points (lower is better). With only two calibration points the accuracy was 0.82 degrees.

It’s hard to tell how the loss of accuracy caused by a shorter calibration process will affect your application. The difference between 0.41 and 0.55 degrees corresponds to 1.5 mm at a distance of 60 cm, which would hardly be noticed in most applications.

My recommendation, considering that your primary user group is children, would be to go with the shorter calibration process. Possibly even fewer than five points.

Q. Why is there no question about correction lenses in the older version?
The question about correction lenses doesn’t affect the calibration as such in any way. It is merely for convenience: first, as a hint or reminder to the users that eyewear actually matters, and that different calibration profiles are needed if you sometimes wear glasses and sometimes wear contacts. Second, it makes it possible to set up user calibration profiles for the different scenarios and switch between them without having to re-calibrate.

Q. Which version of the Installation and Configuration Tools should be used?
We have recently updated the REX development kit and the Gaze SDK to make them better aligned with the new EyeX software. As a part of that effort the Installation and Configuration Tools were made obsolete. You are given the choice to either use EyeX, which will take care of all setup for you, or use the low-level Gaze SDK, which will let you take the full responsibility.

So the question should really be: should I use the low-level Gaze SDK, or the higher-level EyeX SDK?

This question is addressed on, for example, the REX setup page.

[Author]

Posts