A Comprehensive Eye-Tracking System Toward Large FOV HMD.

Area of Science:

• Human-Computer Interaction
• Computer Vision
• Virtual Reality Engineering

Background:

• Eye tracking in immersive virtual reality (VR) head-mounted displays (HMDs) presents significant engineering challenges.
• Constraints include optical layout, illumination, and image acquisition, especially with wide fields of view (FOV).

Purpose of the Study:

• To present an integrated near-eye eye-tracking prototype for immersive VR headsets.
• To develop a real-time software pipeline for stable, end-to-end gaze mapping.
• To demonstrate application-level usability through real-time visualization.

Main Methods:

• Designed a prototype combining customized hardware and a real-time software pipeline.
• Integrated optimized near-eye illumination, image acquisition, pupil detection, and a deep learning gaze-vector estimation model.
• Evaluated the model's performance via ablation studies, comparing an attention-enhanced model with a baseline ResNet-152.

Main Results:

• The attention-enhanced deep learning model achieved an average angular deviation of 1.15°.
• This represents a 61.4% relative reduction in angular deviation compared to the baseline model.
• A real-time visualization demonstrated the integration of pupil diameter, gaze vectors, and blink events.

Conclusions:

• The proposed system offers a cost-effective and reproducible engineering solution for near-eye eye-movement acquisition in VR.
• It provides a technical foundation for future VR interaction design and behavioral analysis.
• The enhanced deep learning model significantly improves gaze estimation accuracy in VR settings.