MLGaze

MLGaze analyzes gaze error patterns in consumer eye tracking systems to identify, classify, and quantify factors that degrade gaze-data accuracy using machine learning.

Key Features:

• Pattern Recognition: Employs machine learning algorithms to identify and classify gaze error patterns arising from different error sources affecting eye trackers.
• Data Processing Tools: Provides Python-based resources for gaze data processing including data augmentation, outlier removal, and generation of gaze features for model training.
• Classification and Modeling: Uses classifiers and regression models to distinguish error sources such as variations in user distance, head pose, and eye-tracker orientation and to model their effects on gaze accuracy.

Scientific Applications:

• Identify Error Sources: Determine specific factors contributing to inaccuracies in gaze tracking.
• Quantify Impacts: Measure and model the extent to which different error sources degrade gaze data quality.
• Predict Variability: Forecast changes in gaze error levels under varying operational conditions.

Methodology:

Collects gaze data from participants under multiple simulated operating conditions, applies preprocessing steps including data augmentation, outlier removal, and gaze-feature extraction, and trains machine learning classifiers and regression models to detect and predict impacts of different error sources on gaze accuracy.