Accuracy and precision of the HTC VIVE PRO eye tracking in head-restrained and head-free conditions

Abstract

Purpose : Eye tracking opens up a variety of novel functions in virtual reality. But, many of those functions, such as gaze-contingent content presentation, require a specific level of eye tracking accuracy. The current study evaluates accuracy and precision of gaze estimates over the whole visual field in head-restrained, as well as head-free conditions.

Methods : Accuracy and precision were tested at 25 sample positions spanning +- 25° horizontally and +- 25° vertically in two separate conditions: head-restrained, as well as head-free. In the first condition, target position was fixed to headset, and subjects were asked to saccade to an appearing target, while keeping the head still. In the head-free condition targets were positioned in a world-fixed coordinate system, and subjects were motivated to move the gaze together with the head in a natural way towards an appearing target and fixate it. In both conditions the displayed targets were presented in a randomized order with 5 repetitions. Eleven subjects were tested in the first condition; ten subjects were tested in the second condition.
In the head-fixed condition accuracy was calculated as the mean offset between the target-eye vector and gaze-target vector. Precision was calculated as the RMS (Root Mean Square) of successive measurements. In the head-free condition, phases with head movement were compared to phases of static head in accuracy and precision within each trial.

Results : Average accuracy of the eye-tracker in the head-fixed condition was 4.16°, SD: 3.23 while the precision had a mean of 2.17°, SD: 0.75. Comparing accuracy and precision horizontally in a one-way ANOVA, revealed that accuracy differed between the centre and periphery (F (4,50) = 3.35, p = 0.02). Under head-movement one-way ANOVA revealed a significant difference in precision between phases with moving head, compared to phases of static head (F (1,18) = 253.03, p < 0.0001; mean_moving: 6.21°, SD: 0.77; mean_static 1.80°, SD: 0.42).

Conclusions : While providing a powerful tool for many novel functionalities the eye tracking of the HTC Pro Eye’s accuracy and precision have to be taken into account in experiment planning, specifically when including tracking in the periphery, or head movements.

This is a 2020 ARVO Annual Meeting abstract.