High-speed and robust infrared-guiding multiuser eye localization system for autostereoscopic display

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Display Technology

Background:

• Autostereoscopic displays require precise eye localization for optimal viewing.
• Existing eye localization methods can be slow and susceptible to environmental factors like lighting and pseudo-faces.

Purpose of the Study:

• To develop a high-speed, robust, and multi-user eye localization system for binocular autostereoscopic displays.
• To improve the accuracy and efficiency of eye tracking in real-world conditions.

Main Methods:

• A system integrating a thermal infrared camera and stereo visible spectral cameras was designed.
• YOLO-V3 neural network was used for face detection in thermal images, effectively filtering pseudo-faces.
• SeetaFace algorithm and template matching were employed for precise eye localization and distance calculation.

Main Results:

• The system achieved an average detection time of 3-8 ms, significantly outperforming traditional methods (86.7%-90.1% improvement).
• The method demonstrated robustness against pseudo-faces and strong ambient light.
• Accurate viewer-to-system distance was calculated using template matching.

Conclusions:

• The proposed infrared-guided eye localization system offers a significant advancement in speed and accuracy for autostereoscopic displays.
• This technology enhances the user experience by enabling more reliable and responsive 3D viewing.
• The system's resilience to environmental challenges makes it suitable for practical, real-world applications.