日本語 日本語

Using the Tobii Pro Glasses 2 eye images to help setup and troubleshoot an eye tracking recording

Tobii Pro Glasses 2 Wearable Eye Trackers Data quality
The Pro Glasses 2 eye tracker uses four cameras (two per eye) to capture images of the eyes. This process is aided by a set of illuminators, that both provide more light for the cameras and create highly visible reflections called glints. The images captured by the camera are used to identify the glints on the cornea and the pupil. This information together with a complex 3 D eye model is then used to estimate the gaze vector and gaze point.

As this process suggests the quality of the gaze point estimation is dependent on the quality of the images captured by the cameras. In this article you will learn how to use the eye images produced by the cameras to help you position correctly the Glasses 2 eye tacker on the participant's head, as well as to troubleshoot the calibration process and recording.

Positioning the glasses on the participant

The quality of the images is ultimately determined by the cameras’ position relative to the eye - a good view of the eyes will produce better gaze point estimations. Fortunately, the Pro Glasses 2 Controller software has two features to help the researcher work out the position of the frame on the participant's head: the Adjustments tool (a.k.a. Track Status), and Eye Images tool. 

 The Adjustments tool provides a simplified view of the positioning of the eyes inside the glasses. The white dots in the image show the location of the pupil in the field of view of the cameras, the large circle. In order for the eyes to be in the field of view of the cameras the dots should stay inside of the circles, and as close to the center of the circles as possible when looking straight ahead.

More importantly, the researcher also has the option to display the eye images recorded by the cameras, when live viewing a recording. These images can be used to inform whether the glasses are positioned correctly, but also help troubleshoot any potential issues with data quality.

You can use the images to check:

  1. if all four cameras have a clear view of the pupils. The pupil should be in plain view and obstructed as little as possible by the eyelids (upper and lower).
  2. that glints are clearly visible. The more glints that are visible, the better, but you should make sure you don’t get additional glints from other light sources (for example, sunlight).
  3. if the eyes are properly illuminated by the illuminators. This can be done by checking if the pupil borders show a good contrast against the iris, i.e. the iris is bright but the pupil is clearly darker.

 

Pro Glasses 2 eye images figure 1

Figure 1.

Figure 1 illustrates a case where the glasses are a bit too low on the nose. Although the pupil is visible by all cameras, most of the illumination ends up on the lower eyelid and cheek, resulting in a too low contrast of the pupil. That is, the iris is also very dark and the borders of the pupil are not easily distinguishable from the iris. Additionally, some glints are covered. This is not a major problem as long as some glints are seen, but to be safe we prefer to have more glints in view if we can. This positioning problem is easily fixable by exchanging the nose piece to one more compatible with your participant.

 

 

Pro Glasses 2 eye images figure 2

Figure 2.

Figure 2 shows the eye images after the nose piece exchange and how this results in better eye images. Most importantly, the eyes are properly illuminated and the pupil can be seen in great contrast. Also, more glints are in view.

 

 

Validating your calibration

During your test it is useful to validate the calibration as soon as you start the recording. This procedure will enable you to check the performance of the eye tracker before collecting the bulk of the data, and document the data quality in the recording, providing you the possibility to refer back to it when needed.

A validation can easily be done by instructing the participant to look at several known positions around the field of view, for example: straight-ahead, up to the left, up to the right, down to the left, down to the right. The offset of the gaze cursor from the known target gives an indication of what offset to expect. If we also consider the eye images, it is possible to see if the error is expected from the calibration, or if it is due to some problem from the eye image. Examples of such errors could be:

  • an obstructed pupil and glints due to partially closed eye lids, because the environment was too bright.
  • an additional and competing glint from an external light source (e.g. sunlight).
  • because the calibration and validation were performed in different light conditions, which changes the pupil size and also its center.

関連記事