To provide reliable results, we have to understand if our predictions about participant’s fixations on the screen were accurate enough.
We are achieving this with a recalibration screen, showing four (4) additional points to the participant of the experiment, after every Eye Tracking session.
Why we display 4 additional points
After every Eye Tracking session, we display four (4) additional points sequentially to validate the accuracy of our predictions.
As you can see in the image above, we display a calibration point per quadrant. After numerous experimental studies, we found out that 4 points are sufficient to determine the accuracy score of our experiments.
The error is the average of all the individual errors of the four (4) recalibration points. We calculate the error of each point with the formula displayed above in the picture. We measure the Euclidean distance between the centre of the calibration (blue) point and the prediction we have made with our algorithms (red point).
Why calibration points are random located in each quadrant
After conducting sufficient research, we found that when Webcam based Eye Trackers are producing low-quality predictions, they tend to make predictions collected in the centre of the web page.
Thus, when platforms re-calibrate the system with fixed points, at the centre of each quadrant, they measure a small error (distance between point and prediction is low), but the truth is that the error is much larger.
By re-calibrating and measuring the system with random point locations, we ensure to measure and deliver the best result to our users.