Automating Areas of Interest Analysis in Mobile Eye Tracking Experiments based on Machine Learning

Area of Science:

• Human-Computer Interaction
• Computer Vision
• Biomedical Engineering

Background:

• Mobile eye tracking analysis requires accurate gaze assignment to Areas of Interest (AOIs).
• Current manual or marker-based methods are time-consuming and impractical for extended recordings involving object interaction.
• Quantitative analysis of long-duration mobile eye tracking studies is challenging.

Purpose of the Study:

• To introduce a novel machine learning algorithm, computational Gaze-Object Mapping (cGOM), for automated gaze data mapping to AOIs.
• To enhance the efficiency and accuracy of mobile eye tracking data analysis.
• To overcome the limitations of existing manual and marker-based gaze mapping techniques.

Main Methods:

• Developed a machine learning algorithm, cGOM, extending Mask R-CNN for object detection, segmentation, and gaze mapping.
• Trained cGOM using a limited dataset of 72 images with 264 object representations.
• Validated cGOM's performance against manual, fixation-by-fixation mapping (ground truth) using True Positive Rate (TPR) and True Negative Rate (TNR).

Main Results:

• cGOM achieved approximately 80% TPR and 85% TNR compared to manual mapping.
• The algorithm demonstrated significant efficiency gains, with a break-even point at 2 hours of total recording time (1 hour of human working time).
• Real-time processing capability was achieved after the training phase, enabling efficient analysis of lengthy recordings.

Conclusions:

• cGOM provides an automated, efficient, and accurate solution for mapping gaze data to AOIs in mobile eye tracking.
• The algorithm significantly reduces the time and effort required for analyzing mobile eye tracking data, especially for studies involving tangible object interaction.
• The developed method facilitates quantitative analysis of extended eye tracking sessions, advancing research in fields utilizing mobile eye tracking.