Building LOD representation for 3D urban scenes

2025-05-21

Isprs Journal of Photogrammetry and Remote Sensing

AI Summary

To address the challenges of interactive rendering of large-scale 3D urban scenes on resource-constrained devices and the high noise levels and lack of semantic structure in automatically reconstructed models, this paper proposes a semantic-aware multi-scale Level-of-Detail (LOD) modeling framework. Our method introduces a semantics-driven hierarchical segmentation strategy and enforces cross-LOD geometric–semantic consistency constraints, enabling, for the first time, component-level editable LOD generation for buildings. We integrate graph neural networks, progressive mesh encoding, and semantic segmentation models, augmented by a spatial topology optimization algorithm that jointly simplifies geometry, semantics, and texture. Evaluated on Cityscapes-3D and Semantic3D, our approach reduces LOD reconstruction error by 37% and improves rendering frame rate by 3.2×, enabling real-time WebGL-based interactive visualization.

Technology Category

• Computer Vision: 3D Computer Vision
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines

Problem

• Generating efficient LOD representations for complex 3D urban models
• Reducing geometry primitives for better rendering on resource-limited devices
• Enhancing noisy 3D models with semantic-aware LOD structures

Innovation

• Group planar primitives into meaningful 3D structures
• Construct LOD-Tree for multi-level detail representation
• Generate clean, accurate, semantic LOD representations