Enhancing Digital Twin Fidelity Through Low-Discrepancy Sequence and Hilbert Curve-Driven Point Cloud Down-Sampling.

Area of Science:

• Computer Vision
• Geometric Modeling
• Data Science

Background:

• Point cloud down-sampling is crucial for digital twin creation, but maintaining geometric fidelity while reducing data volume is challenging.
• Existing methods often struggle with uniform density or require intensive computation for feature detection.

Purpose of the Study:

• To propose a novel point cloud down-sampling method combining Low-Discrepancy Sequences (LDS) and Hilbert curve ordering.
• To preserve both global distribution characteristics and local geometric features in down-sampled point clouds for digital twins.

Main Methods:

• The LDS-Hilbert approach leverages mathematical properties of LDS and space-filling curves for balanced sampling.
• It respects original data density and ensures comprehensive coverage without feature-specific calculations.

Main Results:

• LDS-Hilbert outperformed Simple Random Sampling (SRS), Farthest Point Sampling (FPS), and Voxel Grid Filtering (Voxel) in experiments.
• Significant improvements were observed in parametric model fitting (over 50%) and shape preservation on complex scans (up to 160% better distance metrics).

Conclusions:

• The LDS-Hilbert method offers a practical advance for enhancing digital twin fidelity.
• It achieves superior down-sampling performance without extensive pre-processing or task-specific training data.