Scalable point cloud meshing for image-based large-scale 3D modeling.

Area of Science:

• Computer Vision
• Geometric Modeling
• 3D Reconstruction

Background:

• Image-based 3D modeling reconstructs large-scale scenes, crucial for city-level scenarios.
• Creating watertight mesh models from noisy point clouds is vital for quality reconstruction.
• Existing global Delaunay-based methods struggle with scalability for large datasets.

Purpose of the Study:

• To propose a scalable point-cloud meshing approach for city-scale 3D reconstruction.
• To minimize time consumption and memory usage in large-scale scene reconstruction.
• To overcome the limitations of global optimization methods in handling extensive data.

Main Methods:

• Scene division into overlapping chunks along x and y axes to manage memory constraints.
• Parallel Delaunay-based optimization applied to each chunk for mesh extraction.
• Merging of local meshes with resolution of inconsistencies in overlapping regions.

Main Results:

• Demonstrated scalability on city-scale scenes with millions of points and thousands of images.
• Achieved high accuracy and completeness compared to state-of-the-art methods.
• Efficient processing with minimal time and memory overhead.

Conclusions:

• The proposed chunk-based Delaunay approach offers a scalable solution for large-scale 3D reconstruction.
• Effective for generating high-quality watertight mesh models from complex point clouds.
• Outperforms existing methods in terms of efficiency and handling large datasets.