The quality of collected eye-tracking data is highly participant dependent. For the more-or-less ideal participant (no spectacles or contact lenses, brown eyes, large eye opening, no downward pointing eyelashes, no make up) eye-tracking will generally go smoothly. For some participants however it can be challenging to achieve good precision and/or accuracy.

This page provides a list of undesirable eye-tracking situations you might encounter. Some situations can be remedied by instructing the participant or by adjusting particular eye-tracking settings. Some problems may be harder to tackle.

Unstable eye detection

In case the detection of one or both eyes is unstable or the detected pupil edge of an eye erratically comes and goes, the estimated gaze will be very jittery.

  • Check the iris-pupil separation as described below.

  • If the participant is wearing spectacles make sure they are really clean. See also Clean vs clean spectacles.

  • If the participant is wearing spectacles with anti-reflection or scratch-resistance coating there’s not much you can do to improve detection other than to ask the participant to wear non-coated glasses.

  • Check if the expected pupil size and IPD roughly match the actual pupil size and IPD. See Adjusting eye detection settings.

  • The participant may have a pathological condition causing involuntary eye movements. If this causes detection instability and the condition only exists in one eye, mark that eye as lazy or disable the detection for that eye.

Jumpy CR detection

If the detection of the corneal reflection is unstable it is often due to unexpected additional glints visible in the eye cause by spectacles or tear glints.

  • Make sure spectacles if any are really clean. See also Clean vs clean spectacles.

  • In case of a too teary eye the participant could use a clean tissue to remove the excess tear fluid from the lower eyelid rim.

Noisy eye detection

The proper detection of the pupil and the CR predominantly relies on the quality of the image capture. If the image containing the eyes is noisy, the estimated position of the pupil and CR in the camera image will also be noisy, and so will the derived gaze position.

  • Make sure the participant is sitting at the proper distance from the eye-tracker and the test screen. The eyes should be in focus. See also Preparing the participant.

  • Make sure no daylight (from the sun) enters the test environment as interfering daylight or other light sources will have a negative impact. Close curtains or blinds on outside windows.

  • In case you have reduced the camera shutter time (i.e. the duty-cycle eye-tracker setting), be aware that doing so increases the image noise level.

  • Don’t use extreme camera gain and/or gamma values as this might increase the image noise level.

Low iris-pupil separation (dark iris)

For some participants with bluish irises the gray-level difference between the iris and the pupil (aka the iris-pupil separation) under near infrared light will be too small for the eye-tracker to reliably distinguish the pupil from the iris. The actual iris-pupil separation is shown in eyevec-control (see Iris-pupil separation in the Eyevec-control user manual).

High iris-pupil separation (bright pupilary zone)

The iris-pupil separation (see above) can be rather high for participants with a bright pupilary zone. If the value is very high the eye-tracker might mistake the bright pupilary zone pixels for dust specks or sclera pixels.

One eye higher than the other

If the eyes in the camera image are not (roughly) on the same height then very likely the participant’s face is not perpendicular to the eye-tracker. Assuming the default setup where the eye-tracker is placed in front of the test screen (visually below the active screen area), this means the participant is looking at the test screen with a slight head yaw. It could also be that the eye-tracker is shifted too far from the center of the screen, or it is placed under an angle.

  • Check and if necessary adjust the geometry of the setup.

  • Instruct the participant to properly face the screen, not with head yaw.

One eye darker than the other

This will happen if the participant is not properly centered in front of the eye-tracker and the test screen. Perhaps they are sitting centered with respect to the test screen, but the eye-tracker has been moved to one side too much. One eye appearing darker than the other will also happen under head yaw or if daylight is coming in from the side.

  • Make sure the eye-tracker is centered under the test screen.

  • Assure that the participant when looking straight ahead is (horizontally) looking at the center of the screen.

  • Make sure no daylight (from the sun) enters the test environment. Close curtains or blinds on outside windows.

Baseline measurement reports high noise

See Noisy eye detection above. See also next section.

Baseline/Calibration/validation targets presented twice

Following each target presentation the eye data samples collected during the target presentation are evaluated and the noise level (RMS value of deviation from regression line) of pupil position and CR position are calculated. If the noise levels for the right, left and/or mean eye exceed a threshold (acceptable-pupil-noise-level or acceptable-cr-noise-level) then the target presentation will be repeated. This will occassionally happen. If it happens for more than, say, 10…​20% of the targets then something might be off.

  • Check the points mentioned in Noisy eye detection above.

  • Is the participant not able to sit still (infants) then increase the expected noise level value in the participant settings.

  • In case the participant has a pathological condition causing involuntary movements for one eye mark that eye as lazy or disable the detection for that eye. If the condition is present in both eyes, increase the expected noise level in the participant settings.

  • If possible keep the ambient temperature below 25°C. The hotter the eye-tracker camera is the more noisier the eye images will be.

Bad calibration results

The post-calibration graphics should show a pattern similar to the pattern as presented. If however the participant did not properly look at each target, or points where rejected because of a too high noise level, the calibration results will be unusable. This will be clearly visible in the post-calibration graphics.

  • Instruct the participant to look at each target until it disappears.

  • Make sure the participant has ample time to move their gaze to the target and to fixate on it.

  • For infants use image style targets and don’t use calibration types with more than 5 points (use 2 or 4 points).

Bad/poor validation results

Acceptable validation results obviously requires a valid calibration to begin with. If the calibration is bad it will be impossible to make the validation work.

Post-validation results are presented numerically (in summary) and graphically. The numerical results classifies validation points as good, fair, poor, bad, invalid or skipped. The graphic results shows an indication of the error and standard deviation per target for the right, left and mean eye. If the errors are too large or there are too many errors it might be opportune to redo the validation, to be sure the validation itself went well.

  • Check the points mentioned in Bad calibration results.

  • For infants it is best to skip validation intirely.

  • Repeat the validation if there are one or more points marked poor or bad. Ideally the majority of points is classified as good and a minority as fair. Note, don’t be tempted to repeat validation until all points are good.

  • If necessary relax settings: good-validation-gaze-error, fair-validation-gaze-error, poor-validation-gaze-error, bad-validation-gaze-error.