- Updated: August 25, 2026
- 2 min read
LiDAR‑Based 3D Change Detection at City Scale – A Breakthrough in Urban Mapping
LiDAR‑Based 3D Change Detection at City Scale
Published on ubos.tech
High‑definition 3D city maps are essential for modern urban planning, asset monitoring, and compliance. The recent paper LiDAR‑based 3D Change Detection at City Scale (arXiv:2510.21112v3) introduces an uncertainty‑aware, object‑centric pipeline that dramatically improves change‑detection accuracy on city‑scale LiDAR datasets.
Key Contributions
- Multi‑resolution NDT + point‑to‑plane ICP alignment for robust multi‑temporal registration.
- Per‑point detection confidence derived from registration covariance and surface roughness.
- Semantic and instance segmentation integration for geometry‑based association refinement.
- Class‑constrained bipartite assignment with dummy nodes to handle split‑merge cases.
- Tiled processing that preserves narrow ground changes while keeping memory usage low.
Results on Subiaco, WA
The method achieves 95.3 % accuracy, 90.8 % macro F1, and 82.9 % macro IoU, outperforming the strongest baseline (Triplet KPConv) by 0.3–1.1 percentage points.
Why It Matters for UBOS Tech
Our platform leverages cutting‑edge 3D mapping to provide clients with actionable insights. Incorporating this research enables us to:
- Deliver more reliable change‑detection services for municipal authorities.
- Offer precise volumetric analysis for construction and green‑space monitoring.
- Reduce computational overhead through tiled processing, fitting our scalable cloud architecture.
Read more about our 3D mapping solutions here. For technical details, see the full paper on arXiv.
Stay tuned for upcoming webinars and tutorials on integrating LiDAR change detection into your workflows.
Andrii Bidochko
CTO UBOS
Andrii Bidochko is an AI entrepreneur and researcher focused on AI agents, reinforcement learning, and autonomous systems. He writes about the technologies shaping the future of machine intelligence, from frontier models and agent architectures to real-world AI applications.