Low-density point eating algorithm for surface reconstruction from dense scans.

Area of Science:

• Computer Vision
• Geometric Modeling
• 3D Reconstruction

Background:

• Dense 3D scans often contain noise and varying point densities.
• Existing surface reconstruction methods struggle with low-density overlapping regions.

Purpose of the Study:

• To introduce a new algorithm for accurate surface reconstruction from dense point cloud data.
• To address challenges posed by low-density and poorly scanned regions in 3D scans.

Main Methods:

• Estimation of density maps for each scan with down-weighted boundary densities.
• A 'point eating' strategy to remove low-density overlapping points based on a user threshold.
• Thinning overlapping areas using the moving least-squares operator and weighted averaging of homogeneous points.

Main Results:

• The algorithm successfully extracts smooth yet detailed point set surfaces.
• Reconstructed surfaces closely approximate the ground truth.
• Demonstrated superior performance compared to several advanced surface reconstruction algorithms.

Conclusions:

• The proposed low-density point eating algorithm enhances surface reconstruction accuracy from dense scans.
• It effectively handles noisy and low-density data, producing high-fidelity surfaces.
• Offers a robust solution for 3D surface reconstruction challenges.