Etracker: A Mobile Gaze-Tracking System with Near-Eye Display Based on a Combined Gaze-Tracking Algorithm.

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Biomedical Engineering

Background:

• Eye tracking is crucial for various fields, including psychology, medical diagnosis, and driver assistance.
• Existing gaze-tracking models lack a near-eye display system with both accuracy and user convenience.
• Mobile eye-tracking systems are needed for broader accessibility and real-world applications.

Purpose of the Study:

• To develop a novel mobile gaze-tracking system, Etracker, integrated with a near-eye viewing device.
• To propose a hybrid algorithm combining deep learning and geometric modeling for enhanced gaze tracking.
• To introduce a calibration method that minimizes recalibration frequency for improved user experience.

Main Methods:

• Developed the Etracker prototype, a mobile system with a near-eye display.
• Implemented a combined gaze-tracking algorithm using convolutional neural networks (CNNs) for blink detection and coarse gaze prediction.
• Employed a geometric model for fine-tuning gaze position accuracy.
• Utilized a mean gaze value approach in calibration to mitigate pupil center variations due to nystagmus.

Main Results:

• Achieved 98% eye center detection accuracy in experiments with 26 participants.
• Demonstrated an average gaze accuracy of 0.53° with the Etracker system.
• Operated at a frame rate of 30-60 Hz, ensuring real-time performance.
• The calibration method allows for single-time calibration per user.

Conclusions:

• The Etracker system offers a convenient and accurate solution for mobile human gaze tracking.
• The combined CNN and geometric model algorithm effectively addresses challenges like blinking and nystagmus.
• This technology has significant potential for applications in psychology, medicine, and human-computer interaction.