03-30-2022, 04:56 AM
Why calibration is important
Calibration is performed to provide the EyeLink Host PC with the necessary information it needs to translate the raw positional data —the location of the pupil and corneal reflection (CR)— from the EyeLink Camera into an on-screen gaze location in pixel coordinates. The information below will explain the importance of this and answer the common questions of "How often should I calibrate?", "What makes a calibration "good", and help you understand the critical role calibration plays in determining the spatial accuracy of your data.
How does calibration work?
Calibration works by building a mathematical model that translates raw data from the eye-tracking camera into the specific pixel coordinates of where a person is looking on a screen. Our "How EyeLinks Work" webinar covers the concept in more detail.
Pupil-CR Tracking:
Modern eye trackers use a Pupil-CR (Pupil minus Corneal Reflection) method. For every measurement, the system locates the center of the pupil and the center of the corneal reflection (a small, bright dot from the illuminator). It then calculates the vector between them. This Pupil-CR value is the raw data that changes as the eye rotates.
During calibration, the "D" and "+" symbols you see on the Host PC represent this raw data. Because this data exists in the camera's coordinate space, its pattern may look arbitrary. However, what's critical is that the pattern of these raw data points should mirror the pattern of the calibration targets on the screen.
Mapping Functions:
The main purpose of calibration is to create a mapping function. Think of it like a simple regression model that can predict a person's height from their shoe size. By measuring a few known pairs of shoe sizes and heights, you can create a formula that estimates the height for any shoe size, not just the ones you measured.
Similarly, calibration collects raw Pupil-CR data at a few known screen locations (the targets). It then builds a model to predict the gaze location for the entire screen.
For example, imagine we collect these data pairs for three screen locations:
Using this formula, a new raw data point of 1000 would correctly predict a screen pixel value of 1400.
This is why calibration is so critical: the spatial accuracy of your data is entirely dependent on the quality of this model. A poor calibration will inevitably produce inaccurate gaze data.
What makes a good calibration model?
A good calibration model accurately predicts gaze location across the entire screen. In practice, this means if a participant looks at a non-calibrated point on the screen, the eye tracker reports gaze coordinates that are very close to that true location.
The Importance of Symmetry
A key determinant of a good calibration is the symmetry of the raw data points (the Pupil-CR input). A symmetrical grid of raw data is crucial because it minimizes the impact of non-linearities in the calibration model. While the model can handle minor irregularities, large asymmetries can distort the final gaze map, reducing accuracy.
Common Causes of Asymmetry
The most common cause of an asymmetrical grid is a suboptimal physical setup. This often happens when the monitor is too close to the participant, forcing them to rotate their eyes beyond the camera's trackable range to view the corner targets. This excessive rotation can distort the corneal reflection, producing outlier data points and an unreliable calibration.
For further information about optimal physical set ups, and the importance of making sure that the calibrated area does not exceed the trackable range, please see the following posts:
Calibration Tips
Beyond having an optimal physical set up, the following tips and tricks can all help to ensure good calibrations:
Calibration is performed to provide the EyeLink Host PC with the necessary information it needs to translate the raw positional data —the location of the pupil and corneal reflection (CR)— from the EyeLink Camera into an on-screen gaze location in pixel coordinates. The information below will explain the importance of this and answer the common questions of "How often should I calibrate?", "What makes a calibration "good", and help you understand the critical role calibration plays in determining the spatial accuracy of your data.
How does calibration work?
Calibration works by building a mathematical model that translates raw data from the eye-tracking camera into the specific pixel coordinates of where a person is looking on a screen. Our "How EyeLinks Work" webinar covers the concept in more detail.
Pupil-CR Tracking:
Modern eye trackers use a Pupil-CR (Pupil minus Corneal Reflection) method. For every measurement, the system locates the center of the pupil and the center of the corneal reflection (a small, bright dot from the illuminator). It then calculates the vector between them. This Pupil-CR value is the raw data that changes as the eye rotates.
During calibration, the "D" and "+" symbols you see on the Host PC represent this raw data. Because this data exists in the camera's coordinate space, its pattern may look arbitrary. However, what's critical is that the pattern of these raw data points should mirror the pattern of the calibration targets on the screen.
Mapping Functions:
The main purpose of calibration is to create a mapping function. Think of it like a simple regression model that can predict a person's height from their shoe size. By measuring a few known pairs of shoe sizes and heights, you can create a formula that estimates the height for any shoe size, not just the ones you measured.
Similarly, calibration collects raw Pupil-CR data at a few known screen locations (the targets). It then builds a model to predict the gaze location for the entire screen.
For example, imagine we collect these data pairs for three screen locations:
- Screen X-Coordinates: [80, 960, 1840]
- Raw X-Data: [-8000, -2000, 4000]
Code:
Screen Pixel (x) = Raw Data (x) * 0.147 + 1253Using this formula, a new raw data point of 1000 would correctly predict a screen pixel value of 1400.
Code:
(1000 * 0.147) + 1253 = 147 + 1253 = 1400This is why calibration is so critical: the spatial accuracy of your data is entirely dependent on the quality of this model. A poor calibration will inevitably produce inaccurate gaze data.
What makes a good calibration model?
A good calibration model accurately predicts gaze location across the entire screen. In practice, this means if a participant looks at a non-calibrated point on the screen, the eye tracker reports gaze coordinates that are very close to that true location.
The Importance of Symmetry
A key determinant of a good calibration is the symmetry of the raw data points (the Pupil-CR input). A symmetrical grid of raw data is crucial because it minimizes the impact of non-linearities in the calibration model. While the model can handle minor irregularities, large asymmetries can distort the final gaze map, reducing accuracy.
Common Causes of Asymmetry
The most common cause of an asymmetrical grid is a suboptimal physical setup. This often happens when the monitor is too close to the participant, forcing them to rotate their eyes beyond the camera's trackable range to view the corner targets. This excessive rotation can distort the corneal reflection, producing outlier data points and an unreliable calibration.
For further information about optimal physical set ups, and the importance of making sure that the calibrated area does not exceed the trackable range, please see the following posts:
- EyeLink 1000 Plus Setup / Usage Video Tutorials
- EyeLink Portable Duo Setup / Usage Video Tutorials
- What is the "trackable range" and why is it important?
Calibration Tips
Beyond having an optimal physical set up, the following tips and tricks can all help to ensure good calibrations:
- Choose the appropriate calibration model: Select a model based on your experimental design for the best results. For tasks involving infants/toddlers a HV3 or HV5 calibration are perfect. However for paragraph reading or other scenarios in which maximum accuracy across the entire screen is desired, use the HV9 or HV13 models.
The following thread provides further guidance on which calibration model is appropriate for which situation: Which calibration model should I use?
- Only recalibrate when necessary: Recalibrating without any evidence that the current calibration model is inadequate simply risks replacing a perfectly good calibration model with a potentially inferior one.
The following thread provides further guidance on how often to recalibrate and the use of drift checks. How often should I calibrate?
- Provide clear instructions: For many participants, this will be their first calibration, so it's crucial to provide clear instructions. Tell them to:
- Look directly at the center of each target.
- Keep looking at the target until it disappears.
- Look directly at the center of each target.
- Use Manual Calibration: The default automatic mode is fast, but the manual approach (pressing the spacebar to accept each point) gives you more control and often yields a better calibration. When calibrating manually:
- Don't accept the point too early. Allow the participant time to make small corrective saccades to the center of the target.
- Don't accept the point too late. After a second or two, the participant's eye may begin to drift away from the target.
- Don't accept the point too early. Allow the participant time to make small corrective saccades to the center of the target.
- Use the Backspace keyl: If you accept a fixation point that isn't optimal, you can press backspace to undo it and repeat that target. This can be repeated multiple times to move back multiple targets.
- Use animated calibration targets: When working with participants who may have trouble fixating on static points, such as infants or children, animated targets are more effective. Looming targets with a clear "center of gravity" work especially well to draw and hold their attention.
There are some good examples in this post: Animated Calibration Targets