Eye-tracking Without Screens: Potential Problems and How To Deal With Them

Abstract

Any eye tracking set up has pros and cons and knowing the advantages and disadvantages of using a certain set up and how best to calibrate it is important for researchers choosing a setup for their study. When using an eye tracking setup without screens a lot of relevant questions remain unanswered regarding how to calibrate and use the setup. To answer those questions, a screenless remote eye tracking setup with custom built offline calibration software was used to record the gaze of both eyes of participants while they looked at several calibration and validation points. This gaze data was used to determine the way to calibrate this setup that results in as accurate and precise data as possible and investigate the effect of binocular and monocular data on the accuracy of the results to determine which is better. The constancy and predictability of the parallax error was also studied. The results show that the use of binocular or monocular accuracy led to an improvement in the accuracy of the data, but it cannot be concluded that this difference is significant. Using more calibration points does significantly improve the quality of the data when comparing a 4-point calibration to a 9-point calibration. However, the improvement is so small that in most interaction studies the improvement is not worth the extra time a 9-point calibration takes. The results concerning the distance between calibration points show no difference in data quality when using calibration points placed further apart. The parallax error seems constant enough to manually correct for it with some experience and knowledge about the forward shift of the object from the calibration plane and the location of the stimulus relative to the centre of the screen. It is recommended to test one’s setup extensively before using it in an experiment to determine the best way to calibrate it and use the data and to gain insight into how big of a parallax error one can expect.