Large-Scale Urban Reconstruction with Tensor Clustering and Global Boundary Refinement.

Area of Science:

• Computer Vision
• Computer Graphics
• Geospatial Data Processing

Background:

• Large-scale urban reconstruction from LiDAR data is crucial but challenging.
• Existing methods require labor-intensive, data-dependent user parameters.
• Automating this process is essential for creating functional virtual environments.

Purpose of the Study:

• To develop an automated method for converting large LiDAR point clouds into simplified polygonal 3D models.
• To eliminate the need for user-defined parameters in the reconstruction process.
• To improve the efficiency and accuracy of urban reconstruction.

Main Methods:

• Data is divided into components processed concurrently to extract point metrics.
• Extracted metrics are converted into tensors for clustering.
• A parameter-free agglomerate clustering algorithm segments tensors into geospatial objects.
• A multi-stage boundary refinement process uses global optimization for cluster boundary extraction.

Main Results:

• Successfully segmented geospatial objects like roads and buildings from diverse LiDAR datasets.
• Demonstrated a robust method independent of data-specific parameters.
• Achieved accurate boundary extraction through global optimization.

Conclusions:

• The proposed automated method effectively converts LiDAR point clouds into 3D models without user intervention.
• The parameter-free clustering and boundary refinement offer a significant advancement in urban reconstruction.
• Publicly available source code will facilitate further research and application.