Tracker/Camera Calibration for Accurate Automatic Gaze Annotation of Images and Videos.

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Biomedical Engineering

Background:

• Appearance-based gaze tracking requires extensive, accurately annotated training data.
• Current annotation methods rely on subjects fixating on known points, limiting naturalistic data collection.
• Ground truth gaze direction is crucial for training robust gaze estimation models.

Purpose of the Study:

• To develop a method for generating gaze annotations in natural settings without requiring target point fixation.
• To enable the use of infrared (IR) gaze trackers for creating high-quality gaze annotation datasets.
• To simplify and improve the efficiency of data acquisition for gaze tracking research.

Main Methods:

• Proposed a novel approach using an IR gaze tracker to generate gaze annotations.
• Developed a geometric calibration procedure between the IR gaze tracker and a camera using the PnP algorithm.
• Demonstrated the calibration procedure's effectiveness in aligning IR gaze data with the camera's reference frame.

Main Results:

• Successfully generated gaze annotations in unconstrained, naturalistic settings.
• The proposed calibration method allows IR gaze data to be accurately expressed in the camera's coordinate system.
• The generated annotations provide a full characterization of gaze direction for training.

Conclusions:

• The proposed IR gaze tracker-based annotation method significantly enhances data collection for gaze tracking.
• This technique removes the artificial constraint of target fixation, leading to more realistic training data.
• The PnP-based calibration offers a straightforward and effective solution for integrating IR gaze data into camera-based systems.