A Local Graph-Based Structure for Processing Gigantic Aggregated 3D Point Clouds.

Area of Science:

• Computer Vision
• 3D Data Processing
• Computational Geometry

Background:

• Gigantic point cloud datasets pose significant scientific challenges due to their sheer volume.
• Existing methods struggle with memory limitations when processing aggregated scans.
• Accurate representation and simplification of large-scale 3D scenes are crucial for analysis and visualization.

Purpose of the Study:

• To present an original workflow for structuring aggregated point clouds from multiple scans.
• To develop a memory-efficient method for handling large-scale 3D data.
• To demonstrate the relevance of the proposed structure for data resampling and visualization.

Main Methods:

• A novel representation based on a set of local graphs derived from depth maps.
• Connecting graphs via overlapping scan areas, managing redundant points for consistency.
• Developing a resampling algorithm leveraging the structured point cloud representation.

Main Results:

• Achieved a piecewise and globally consistent structure for sampled point clouds.
• Demonstrated the workflow's capability to structure aggregated point clouds regardless of acquisition number or size.
• Successfully implemented a resampling algorithm capable of handling billions of points on standard computers.

Conclusions:

• The proposed workflow offers an efficient solution for structuring large point clouds, overcoming memory constraints.
• The developed structure is highly relevant for managing and processing massive 3D datasets.
• The resampling algorithm facilitates simplification and visualization of extremely large-scale scenes like urban environments and historical sites.