Strategies for enhancing automatic fixation detection in head-mounted eye tracking.

Area of Science:

• Computer Vision
• Neuroscience
• Human-Computer Interaction

Background:

• Accurate fixation detection is essential for analyzing visual information, especially in eye-tracking studies.
• Standard algorithms perform well in static conditions but struggle with head motion.
• Head-mounted eye trackers enable naturalistic movement but complicate fixation detection.

Purpose of the Study:

• To enhance existing fixation detection algorithms for robust performance in dynamic, real-world scenarios using head-mounted eye trackers.
• To develop strategies that differentiate gaze stabilization from non-fixational gaze shifts during head movement.

Main Methods:

• Proposed optic-flow-based compensation for stabilizing eye movements during head motion.
• Implemented adaptive algorithm sensitivity adjustment based on head-motion intensity.
• Utilized a new, publicly available hand-labeled dataset from Pupil Invisible glasses for validation.

Main Results:

• Individual and combined strategies were evaluated for their contribution to fixation detection accuracy.
• The combined strategies significantly improved standard thresholding algorithms.
• The proposed methods outperformed previous approaches for head-mounted eye tracking.

Conclusions:

• The developed strategies enable robust fixation detection in dynamic environments using head-mounted eye trackers.
• This advancement is critical for analyzing eye movements in naturalistic settings.
• The findings advance the field of eye-tracking analysis for real-world applications.