Point cloud feature edge point curve fitting

Area of Science:

• Computer Vision
• Geometric Modeling
• Computational Geometry

Background:

• Structured light scanning generates edge point clouds prone to scattering noise and non-uniform sampling.
• Traditional discrete fitting methods struggle with noisy and irregularly sampled data, limiting accuracy in 3D reconstruction.

Purpose of the Study:

• To develop a robust, one-stage framework for direct parametric reconstruction of 3D edges from noisy point clouds.
• To bypass intermediate discrete representations and directly predict accurate geometric primitives.

Main Methods:

• A one-stage parametric reconstruction framework integrating mask-driven dynamic control point regression.
• A differentiable geometric constraint solver for direct prediction of Bézier curves, lines, and arcs.
• Directly fitting parametric primitives from scattered points, avoiding discrete representations.

Main Results:

• Achieved a Chamfer Distance of 0.0038 and a Hausdorff Distance of 0.0764 on the ABC dataset.
• Demonstrated high robustness against scanning noise and non-uniform sampling.
• Preliminary tests on real-world sensors confirmed method robustness.

Conclusions:

• The proposed framework offers a robust solution for 3D edge parameterization from noisy point clouds.
• Direct parametric prediction bypasses limitations of discrete fitting, enhancing accuracy for reverse engineering.
• The method shows potential for real-world applications in 3D scanning and modeling.