Data-driven approaches to unrestricted gaze-tracking benefit from saccade filtering

Unrestricted gaze tracking that allows for head and body movements can enable us to understand interactive gaze behavior with large-scale visualizations. Approaches that support this, by simultaneously recording eye- and user-movements, can either be based on geometric or data-driven regression mode...

Full description

Saved in:
Bibliographic Details
Published in2016 IEEE Second Workshop on Eye Tracking and Visualization (ETVIS) pp. 1 - 5
Main Authors Flad, Nina, Ditz, Jonas C., Schmidt, Albrecht, Bulthoff, Heinrich H., Chuang, Lewis L.
Format Conference Proceeding
LanguageEnglish
Published IEEE 23.10.2016
Subjects
Biology
Cameras
Data models
Magnetic heads
Tracking
Training
Training data
Online AccessGet full text

Cover

More Information
Summary:Unrestricted gaze tracking that allows for head and body movements can enable us to understand interactive gaze behavior with large-scale visualizations. Approaches that support this, by simultaneously recording eye- and user-movements, can either be based on geometric or data-driven regression models. A data-driven approach can be implemented more flexibly but its performance can suffer with poor quality training data. In this paper, we introduce a pre-processing procedure to remove training data for periods when the gaze is not fixating the presented target stimuli. Our procedure is based on a velocity-based filter for rapid eye-movements (i.e., saccades). Our results show that this additional procedure improved the accuracy of our unrestricted gaze-tracking model by as much as 56 %. Future improvements to data-driven approaches for unrestricted gaze-tracking are proposed, in order to allow for more complex dynamic visualizations.
DOI:10.1109/ETVIS.2016.7851156