Abstract

Eye tracking is an increasingly popular method for interacting with augmented reality (AR) and measuring user attention. However, implementing and evaluating eye tracking across multiple platforms and use cases can be challenging due to the lack of standardized metrics and measurements. Additionally, existing calibration methods and accuracy measurements often do not account for the common scenarios of walking and scanning in mobile AR settings. To test and compare different eye tracking devices on various AR tasks and metrics, we developed EyeTTS, an eye tracking test suite specifically designed for scenarios involving head movement and locomotion in AR. We conducted user studies on the Magic Leap (n=36, 1 trial per task) and HoloLens 2 (n=54, 2 trials per task) devices to collect data and compare the precision and accuracy of each headset under different movement and reference frame conditions. Additionally, we performed post-hoc recalibration of the eye gaze data using the tasks in our test suite in order to study the impact of different calibration tasks on accuracy and compare them with the inbuilt calibration provided by the head-mounted displays. Our analysis revealed lower eye tracking accuracy while walking, significant differences between the two tested headsets, and improvements to eye tracking performance while walking when a calibration technique involving stipulated head motion is employed.