Accuracy of high-end and self-build eye-tracking systems

Abstract

Eye tracking is a promising technology for human-computer interactions, which is however rarely used in practical applications. We argue that the main drawback of the contemporary eye trackers is their limited accuracy. There is no standard way of specifying this accuracy what leads to underestimating the accuracy error by eye tracker manufacturers. In this work we perform a subjective perceptual experiment measuring the accuracy of two typical eye trackers: a commercial corneal reflection-based device mounted under a display and a head-mounted do-it-yourself device of our construction. During the experiment, various conditions are taken into consideration including viewing angle, human traits, visual fatigue, etc. The results indicate that eye tracker accuracy is observer-referred and measured gaze directions exhibit a large variance. Interestingly, the perceptually measured accuracy of the low-cost do-it-yourself device is close to the accuracy of the professional device.