Computational Large Field-of-View RGB-D Integral Imaging System

Area of Science:

• Optics and Photonics
• Computer Vision
• Image Processing

Background:

• Integral imaging offers real-time 3D displays but traditional methods use physical lens arrays causing distortions.
• Computational integral imaging with virtual lens arrays provides flexibility but struggles with large-scale images, resulting in limited fields of view.
• Existing systems often require additional devices for depth information, hindering practical application.

Purpose of the Study:

• To develop a single image-based computational integral imaging pickup system for a large field of view in real time.
• To overcome the limitations of small virtual lens arrays and the need for supplementary depth devices in previous computational integral imaging methods.
• To enhance the flexibility and applicability of integral imaging for real-time 3D display technologies.

Main Methods:

• Deep learning-based automatic depth map estimation from a single RGB input image.
• Implementation of a hierarchical integral imaging system for capturing a large field of view in real time.
• Application of an inpainting method for post-processing and optimizing visualization of image areas with pickup failures.

Main Results:

• The proposed system successfully estimates depth information from RGB images without external sensors.
• Achieved a large field of view in real-time integral imaging pickup.
• Demonstrated robust performance through quantitative and qualitative experimental results, including optimized visualization of failed areas.

Conclusions:

• The single image-based computational RGB-D integral imaging system effectively addresses the limitations of prior methods.
• The system provides a flexible, real-time solution for large field-of-view 3D imaging without supplementary devices.
• The approach shows robustness and potential for advanced real-time 3D display applications.