3D reconstruction of concave specular objects via single-pixel imaging and implicit representations

Area of Science:

• Computer Vision
• Optical Engineering
• Computational Geometry

Background:

• 3D reconstruction of specular objects is difficult due to complex geometry and reflections.
• Existing methods struggle with accurate light-path correspondence extraction.
• Applications include industrial inspection, optical calibration, and augmented reality.

Purpose of the Study:

• To develop a robust framework for accurate 3D reconstruction of concave specular objects.
• To overcome limitations of traditional methods in handling complex reflective surfaces.
• To enable precise geometry recovery for advanced computer vision tasks.

Main Methods:

• A two-stage framework combining Fourier-based single-pixel imaging and implicit surface representation.
• Decomposition of reflected light into single and multiple reflection components.
• Using point-to-ray correspondences as physical priors to constrain Signed Distance Function (SDF)-based reconstruction.

Main Results:

• Achieved reliable point-to-ray correspondences by analyzing captured reflected light.
• Enabled precise geometry recovery through SDF-based reconstruction constrained by physical priors.
• Demonstrated superior robustness and adaptability compared to traditional methods on synthetic and real data.

Conclusions:

• The proposed framework effectively addresses the challenges of 3D reconstruction for specular objects.
• The combination of Fourier imaging and implicit surfaces offers a significant advancement in 3D vision.
• The method shows high potential for applications requiring accurate 3D models of reflective surfaces.