Performing a calibration and validation in Pro Lab

Tobii Pro Lab Screen Based Eye Trackers Screen based projects

Learn how to perform a calibration and calibration validation with Tobii Pro Lab.

Why do you need to calibrate the eye tracker

The human eye shows a natural individual variation in its shape and geometry. For example, the exact location of the fovea, the area responsible for the part of the visual field that we perceive in focus and full color, varies from individual to individual. Tobii Pro Eye Trackers, use an individual calibration procedure to optimize its gaze estimation algorithms (i.e. the 3D eye model) and account for this variation. The end result of this optimization is a fully-customized and more accurate gaze point calculation. Additionally, some Tobii Pro eye trackers use the calibration procedure to select the detection mode (e.g. dark or bright pupil tracking) that provides the most accurate data. 

The visual axis

Figure 1. The geometry of the visual axis of the eye, optical axis, fovea, center of the pupil (CP) and gaze point. One of the objectives of the calibration is to finetune the geometry to get as close as possible to the true location of the fovea the participant. 

Pro Lab calibration procedure

The default calibration procedure in Pro Lab can be divided into three distinct phases:

  1. A data collection phase, where participants are instructed to look at a certain number of targets, displayed sequentially on the screen.
  2. An optimization phase, where part of the collected data is used to optimize the 3D eye model.
  3. A validation phase, where the remaining data is used to test the new 3D eye model configuration and data quality measures are reported.

Only the first phase of the calibration procedure is clearly visible to the operator and the participant. The remaining two other phases run "under the hood" of the system, and only the data quality results are displayed at the end of the calibration procedure.

During the data collection phase, Pro Lab presents two sets of targets. The first set, the calibration targets, is used during the optimization phase, while the second set, the validation targets, is used during the validation phase. The number of calibration targets can be changed by the test operator (i.e. 3, 5 or 9 targets). The larger the number of calibration targets, the more information is provided to the system to optimize the 3D eye model. The validation targets are fixed to a number of 4, and these targets are displayed in fixed locations, that do not match the calibration targets' locations.

In the optimization phase, the first set of calibration targets obtained during the data collection phase are first mapped to the screen, using, either, a standard configuration of the 3D eye model or the last configuration saved to the eye tracker (e.g. the previous test participant). The system then calculates the position and distribution relationship between the mapped data and the location of the calibration targets. In the next step, the eye tracker changes parameters in the 3D eye model (e.g. the angle between the visual axis and the optical axis) and runs the data through it again, calculating a new set of gaze points and its distance and distribution relative to the targets. It then compares the results of the previous configuration with the new one. If the new configuration provides better results globally, across the different calibration targets, then the previous one is discarded. This configuration tweaking continues until there is no more improvement in the results, at this point the 3D eye model is considered optimized. The results of the optimization procedure will be displayed as a spread of gaze points in the calibration feedback diagram.

In the validation phase, the optimized 3D eye model is used to map the remaining data, collected during the presentation of the validation targets in the data collection phase. The system then calculates the position and distribution relationship between the mapped data and the location of the validation targets and reports the results as the overall accuracy (average distance to targets), precision (standard deviation of the distance to targets) and the number of valid eye tracking samples.

At the end of the calibration procedure, Pro Lab will display both an illustration of the calibration and validation errors and a table with the numerical results of the validation. The operator must then decide to reject or approve the calibration according to the calibration validation feedback. When the operator approves the calibration, the new 3D eye model configuration is then saved to the eye tracker. This configuration is used to calculate the subsequent gaze points during the recording session until a new calibration is performed.

Why are validation targets included in the calibration?

If you choose to report the data quality using the data collected to optimize the 3D eye model, you lose predictive power when estimating the accuracy and precision error of the recording. This is because the 3D eye model is being fine-tuned to that particular data set data and is likely to report better performance than when a new data set is used, which is the case when you perform a recording. As a consequence, it is best practice to test the new configuration of the 3D eye model with a new data set (i.e. validation targets). These values can then be used in your report as an estimate of the data quality.   

Performing a calibration

Now that you have a basic understanding of why you need to calibrate the eye tracker, let's see how you can set up and perform a calibration in Pro Lab.

To set up the calibration procedure:

  1. Open a screen project type.
  2. Click on the Design tab. All new projects start with a Calibration element on the Timeline.
  3. Click the Calibration element to open the calibration properties on the right-side panel.
  4. Toggle the Validate calibration selector to enable/disable* the presentation of the validation targets. 
  5. Select the background color.
  6. Select the number of calibration targets: 3, 5 or 9. The number of validation targets is fixed and equal to 4.
  7. Choose the calibration type: Timed (the points are displayed automatically for a pre-set period of time) or Manual (the operator decides when to collect the data, by pressing a key on the keyboard). 
  8. Choose the type of calibration target: either a Point, default target, or a custom Video type of target (you can choose among the existing videos or import your own).
  9. Once you have set up your calibration and study design you will be ready to perform your first recording.

*If you disable the presentation of the validation targets, after calibration, Pro Lab will display both a diagram and a table with the numerical results of the calibration instead. These results represent the residual error after the 3D eye model has been optimized. Thus, you will only be able to use this information as a first step to help you troubleshoot tracking issues and perform a new calibration or decide whether to exclude test participants with sub-optimal results.  

To start a calibration in a screen based project in Pro Lab:

  1. Open your screen based project.
  2. Open the Record tab.
  3. Click the Start Recording button on the presentation screen.
  4. Adjust the position of the participant using the track box feedback (read Calibration positioning tips below) and provide the calibration instructions to the participant.
  5. Click Start calibration. 
  6. After the targets have been displayed and the data has been collected, the calibration and validation results are shown on the screen. Decide whether:
    • Continue the recording and proceed to the stimulus presentation, click Continue. 
    • Deal with calibration problems in specific target locations, by selecting the targets and then select Recalibrate n points to recalibrate the participant. 
    • Cancel the recording, press Escape (or Shift + Esc if that was selected). 

To review and export your calibration and validation results:

  1. Select a participant’s row in Project Overview.
  2. Click the down arrow to see more information. If there is a calibration or validation result, you’ll see the estimated accuracy and precision values. These results represent the total average results across all targets. 
  3. To save the image, click the Save image to file button at the bottom right and navigate to where you want to save the .png file.
  4. To export all the results, click on Export Calibration data on the heading of the recording list (Top-right corner). The export file contains the data of all recordings in the recordings list. The values in the export file are average values for the whole screen.

Calibration procedure feedback

Use the information displayed in the calibration feedback window as the first step to help you:

  • Troubleshoot tracking issues and perform a new calibration.
  • Decide whether to exclude participants with sub-optimal results.
  • Report an estimate of the accuracy and precision error for the recording.


The top section of this window displays the graphics results, were each calibration target location is represented by white crosshairs, the validation targets by purple “Xs”, and the average recorded gaze point by an orange dot. A faint orange line shows the average length and direction of the gaze estimation error, and a circle illustrates the standard deviation (radius). You can check the accuracy and precision errors for individual target locations by hovering the mouse over the targets.

On the bottom of the calibration feedback window, you will find a table containing the total average accuracy and precision values from all calibration targets. The accuracy and precision values are displayed in degrees, pixels and in millimeters. Values for the individual eyes can be shown by selecting the Left and right eyes results option in the View options tick box.

 Pro Lab calibration feedback window

Figure 2. Calibration and validation feedback in Pro Lab. The white crosshairs represent the calibration target locations and the purple crosses the validation target locations.

Calibration procedure feedback without validation targets

If you have disabled the Calibration Validation procedure, after calibration, Pro Lab will still display both a diagram and a table with numerical calibration results. However, instead of validation results, these values represent the residual error after the 3D eye model algorithm has been optimized.

Use the information displayed in the calibration feedback window as the first step to help you:

  • Troubleshoot tracking issues and perform a new calibration.
  • Decide whether to exclude participants with sub-optimal results.

Note that the calibration results without collecting the data from validation targets, loose predictive power to accurately estimate the gaze mapping error of your recording. In most cases, the reported error will be lower than if you collect new data and process it with the 3D eye model compared to using the feedback from the optimization process. As a result, the calibration results are not optimal if you wish to report the data quality of the recording.  

On the top section of the calibration feedback window, an illustration of the gaze point estimation error is displayed, where each calibration target location is represented by a white crosshair and the estimated gaze points as orange dots. A faint orange line shows the average length and direction of the gaze estimation error. If you hover with your mouse over each calibration target, the accuracy and precision (standard deviation) values will be displayed.

On the bottom of the calibration feedback window, you will find a table containing the total average accuracy and precision values from all calibration targets. The accuracy and precision values are displayed in degrees, pixels and in millimeters. Values for the individual eyes can be shown by selecting the Left and right eyes results option in the View options tick box.

Best Practices

In this section, we have collected a few practical tips and tricks to enable you to optimize your calibration procedure and run a successful calibration.

Calibration design

  • For good performance, choose a calibration background that matches the color and brightness of your stimuli.
  • Use more calibration targets if you can, but remember that there is a trade-off between test participants focus on the procedure and the number of targets.
  • Don't disable the validation targets.

Participant positioning

  • Before the calibration and recording, make sure the participant is in a comfortable posture they can maintain for the duration of the recording.
  • The participants should be facing the front of the eye tracker and screen.
  • The eyes of your participant should be in the center of the track box. The eyes should be in the center of the track status feedback window. The participant should be at the optimal distance to the eye tracker (Please check your eye tracker user manual for the optimal distance recommendation).
  • Make sure the participant can comfortably read the text on the screen, if any, and reach the keyboard and mouse.

Calibration feedback

  • Determine the size of your stimuli items and areas of interest (AOIs) in pixels. Pay particular attention to the size of the smallest items, AOIs, and the space between them.
  • After calibration, diagrams will be presented showing error vectors for each target location (and eye if enabled), and numerical estimation of accuracy, precision and recorded valid samples.
  • Combine the information from the two previous bullet points to decide if you should: a) begin recording using the current calibration or b) adjust the eye tracker or participant and then recalibrate
  • If it is a pilot and the offset is large (for the study), then consider redesigning the stimuli and place the items and related AOIs further apart.
  • Dealing with calibration problems:
    • Missing calibration in a specific target location: Click the missing target location on the illustration and recalibrate.
    • Large offsets (Accuracy): Tell participant not to guess where the target is but rather just follow it.
    • High noise (Precision): Clean glasses and eye tracker, open eyes wider, pin back hair.

Related Articles