AHC-NeRF: Autonomous, High-Quality Neural Reconstruction of Two-Layer Complex Nested Transparent Objects.

Area of Science:

• Computer Vision
• Computer Graphics
• Computational Imaging

Background:

• Reconstructing transparent objects is challenging due to light refraction and reflection.
• Existing methods struggle with complex nested objects and require tedious view-capture strategies.

Purpose of the Study:

• To propose AHC-NeRF, an autonomous, high-quality neural SDF-based framework for reconstructing two-layer complex nested transparent objects.
• To overcome limitations of existing refraction-tracing methods and improve reconstruction precision.

Main Methods:

• Combines neural SDF with single-pixel imaging (SPI) for reflection-based reconstruction of outer and inner surfaces.
• Introduces an adaptive SPI method for accelerated acquisition of point-pair priors.
• Employs a novel view-planning strategy for progressive viewpoint selection based on information gain.

Main Results:

• AHC-NeRF achieves high-quality reconstruction of both outer and inner surfaces of nested transparent objects.
• The adaptive SPI method accelerates prior acquisition by 1-2 orders of magnitude.
• The proposed view-planning strategy enhances surface reconstruction quality.

Conclusions:

• AHC-NeRF demonstrates superior performance compared to state-of-the-art methods on synthetic and real-world datasets.
• The framework offers an effective solution for reconstructing complex transparent objects autonomously.