EasyEyes: Crowded Dynamic Fixation for Online Psychophysics

Fengping Hu , Joyce Yi Xin Chen , Denis G Pelli

Biorxiv : the Preprint Server for Biology +

|

March 10, 2025

View abstract on PubMed

Summary

Crowded dynamic fixation, a new online vision testing method, significantly improves eye gaze accuracy by using visual clutter. This technique enhances participant fixation, making online vision tests more reliable and precise.

Area Of Science

Ophthalmology
Computer Vision
Human-Computer Interaction

Background

Online vision testing offers efficient data collection but requires precise gaze tracking.
Traditional webcam-based gaze tracking lacks sufficient accuracy for online studies.
Existing dynamic fixation tasks reduce, but do not eliminate, fixation errors.

Purpose Of The Study

To introduce and evaluate a novel 'crowded dynamic fixation' task for enhancing gaze accuracy in online vision testing.
To assess the effectiveness of crowding in further improving fixation compared to previous methods.
To determine if crowded dynamic fixation can eliminate peeking behavior during peripheral threshold measurements.

Main Methods

Developed a 'crowded dynamic fixation' task incorporating visual clutter around a moving target.
Assessed fixation accuracy by measuring RMS gaze error and peeking frequency during peripheral threshold measurements.
Compared the crowded dynamic fixation task against stationary and dynamic fixation tasks.

Main Results

Crowded dynamic fixation reduced RMS gaze error by 47% compared to stationary fixation.
Peeking occurred in 0% of trials with crowded dynamic fixation, versus 4% for dynamic and 9% for stationary fixation.
The method eliminated implausibly low peripheral thresholds, suggesting improved fixation accuracy.

Conclusions

Crowded dynamic fixation provides accurate gaze control for online vision testing.
This method effectively leverages crowding to minimize fixation errors and peeking.
The technique holds promise for improving the reliability of remote vision assessment tools.