Compensating for camera translation in video eye-movement recordings by tracking a representative landmark selected...

Area of Science:

• Oculomotor research
• Vestibular research
• Biomedical engineering

Background:

• Eye movement data acquisition commonly uses cameras.
• Unwanted facial motion contaminates data, especially in dynamic environments.
• Accurate eye tracking requires compensating for camera-device relative motion.

Purpose of the Study:

• To develop a method for estimating camera motion relative to a deformable facial surface.
• To automatically select and track a region of interest (ROI) for accurate motion estimation.
• To improve the precision of eye movement data in challenging experimental conditions.

Main Methods:

• Utilizing a genetic algorithm to select optimal ROIs for tracking.
• Employing co-correlation to predict and select accurate ROIs.
• Implementing ROI recombination to generate improved tracking candidates.
• Tracking a selected facial ROI to measure vertical camera translation.

Main Results:

• The genetic algorithm successfully identifies ROIs for accurate camera motion estimation.
• Co-correlation effectively predicts ROI accuracy, enabling optimal ROI selection.
• The developed method achieves an average accuracy of 0.75 degrees in estimating eye-gaze direction.
• Correction significantly reduces errors caused by camera motion (e.g., 11 degrees error reduced).

Conclusions:

• The genetic algorithm-based ROI selection method robustly estimates camera motion.
• Accurate eye-gaze estimation is achievable even with significant camera translations.
• This technique is crucial for reliable oculomotor and vestibular research in dynamic settings.