Texture Illumination Separation for Single-Shot Structured Light Reconstruction.

Area of Science:

• Computer Vision
• 3D Reconstruction
• Computational Imaging

Background:

• Active illumination methods for 3D shape estimation face a trade-off between acquisition speed and reconstruction density.
• Single-shot methods are fast but yield sparse reconstructions and struggle with textured surfaces, while multi-shot methods require static scenes.

Purpose of the Study:

• To develop a single-shot approach for dense 3D shape reconstruction of highly textured objects under active illumination.
• To overcome limitations of existing methods regarding speed, scene dynamics, and reconstruction density.

Main Methods:

• An image decomposition scheme to separate illumination and texture components from mixed-appearance images.
• Handling of three mixed-appearance scenarios: single-projector illumination on textured surfaces, multi-projector illumination on textureless surfaces, and combined effects.
• Computation of per-pixel warps using illumination patterns and a texture template, obtained via interleaved all-white patterns.

Main Results:

• Accurate per-pixel warps computed even with infrequent pattern interleaving and significant object deformation.
• Detailed 3D shape reconstruction and dense motion tracking of textured surfaces achieved.
• Validation on synthetic and real data, demonstrating robustness to non-rigid surface deformations.

Conclusions:

• The proposed single-shot method enables dense 3D shape reconstruction for dynamic, textured scenes.
• The image decomposition technique effectively recovers scene and illumination information from mixed appearances.
• This approach advances active illumination techniques for complex object reconstruction.