Robust and Fast Normal Mollification via Consistent Neighborhood Reconstruction for Unorganized Point Clouds

Area of Science:

• Computer Vision
• Geometric Processing
• 3D Data Analysis

Background:

• Accurate normal estimation is crucial for 3D point cloud processing.
• Existing methods struggle with sharp features and non-uniform sampling.
• Robustness in normal estimation is essential for reliable downstream tasks.

Purpose of the Study:

• To introduce a novel, robust normal estimation method for point cloud data.
• To effectively handle both smooth and sharp geometric features.
• To improve normal estimation accuracy in complex and non-uniformly sampled scenes.

Main Methods:

• Neighborhood recognition integrated into normal mollification.
• Utilizing a normal estimator of robust location (NERL) for smooth regions.
• Employing robust feature point recognition, Gaussian maps, and clustering for sharp features.
• Implementing a two-stage normal mollification process, including a residual-based second stage.

Main Results:

• The proposed method accurately estimates normals for both smooth and sharp features.
• Demonstrated robustness in handling non-uniform sampling and complex point cloud data.
• Experimental validation on synthetic and real-world datasets showed superior performance compared to state-of-the-art methods.

Conclusions:

• The developed method offers a significant advancement in robust normal estimation for point clouds.
• It provides reliable surface normal computation, crucial for various 3D applications.
• The approach effectively addresses limitations of existing normal estimation techniques.