A Cluster-Based 3D Reconstruction System for Large-Scale Scenes

Area of Science:

• Computer Vision
• 3D Reconstruction
• Geospatial Technology

Background:

• Realistic large-scale 3D scene reconstruction from aerial data is crucial for smart cities, surveying, and military applications.
• Current methods face significant challenges due to massive scene scale and data volume, hindering rapid reconstruction.
• Existing pipelines struggle with computational bottlenecks for processing extensive aerial imagery.

Purpose of the Study:

• To develop a professional system for efficient large-scale 3D reconstruction.
• To overcome the limitations of current state-of-the-art 3D reconstruction pipelines.
• To enhance the speed and quality of 3D scene modeling from aerial images.

Main Methods:

• Sparse point-cloud reconstruction: Utilizes a clustered camera graph for local structure-from-motion (SFM) and global camera pose optimization.
• Dense point-cloud reconstruction: Employs checkerboard grid sampling for pixel-level adjacency decoupling and normalized cross-correlation (NCC) for optimal depth estimation.
• Mesh reconstruction: Incorporates feature-preserving simplification, Laplace smoothing, and detail recovery for improved mesh quality.

Main Results:

• The developed system effectively improves the reconstruction speed of large-scale 3D scenes.
• Integration of local SFM with global alignment enhances camera pose accuracy.
• Advanced mesh processing techniques lead to higher quality 3D models.

Conclusions:

• The proposed system offers a viable solution for accelerating large-scale 3D reconstruction.
• The methodology successfully addresses computational challenges in processing vast aerial datasets.
• This work contributes to advancing the practical applications of 3D scene modeling in various fields.