Eye movement analysis for real-world settings using segmented linear regression.

Area of Science:

• Neuroscience and Cognitive Science
• Computer Vision and Signal Processing

Background:

• Eye movement analysis is crucial for understanding perception, cognition, and behavior.
• Real-world gaze data is often noisy, hindering precise analysis of slow transient changes.
• Existing methods struggle with low-precision devices and uncontrolled environments.

Purpose of the Study:

• To propose an Improved Naive Segmented linear regression (INSLR) algorithm for robust eye movement analysis.
• To address challenges of gaze denoising and prediction errors in uncontrolled settings.
• To enhance the accuracy of oculomotor event classification and attention variation detection.

Main Methods:

• Developed INSLR, an approach approximating gaze coordinates with a piecewise linear function (PLF).
• Employed a hypotheses clustering algorithm with least square fit score and k-means distance for hypothesis estimation.
• Filtered hypotheses using a pre-defined threshold to refine gaze signal approximation.

Main Results:

• INSLR consistently outperformed the baseline NSLR in denoising noisy eye movement datasets.
• Minimized gaze prediction errors from low-precision devices (webcams) in 71.1% of samples.
• Validated improved denoising through enhanced accuracy of the NSLR-HMM oculomotor event classifier.

Conclusions:

• INSLR offers a significant advancement for eye movement analysis in real-world, uncontrolled settings.
• The method effectively denoises gaze data and reduces prediction errors, even with cost-effective hardware.
• INSLR improves the detection of subtle changes in attention and oculomotor events.