Uncertainty-Aware LiDAR Change Detection for 3D City Maps
A new research paper on arXiv (2510.21112) proposes an uncertainty-aware, object-centric method for city-scale LiDAR-based 3D change detection. The method addresses limitations of conventional Digital Surface Model (DSM) and image differencing, which are sensitive to vertical bias and viewpoint mismatch, and of point cloud or voxel models that require large memory and perfect alignment. The approach aligns multi-temporal data using multi-resolution Normal Distributions Transform (NDT) and point-to-plane Iterative Closest Point (ICP), normalizes elevation, and computes per-point detection levels from registration covariance and surface roughness. Geometry-based associations are refined by semantic and instance segmentation. The paper was announced as a replace-cross on arXiv, indicating a revision. The method aims to support city planning, municipal compliance, map maintenance, and asset monitoring, including built structures and urban greenery.
Key facts
- Paper arXiv:2510.21112 proposes a new LiDAR-based change detection method.
- The method is uncertainty-aware and object-centric.
- It uses multi-resolution NDT and point-to-plane ICP for alignment.
- It normalizes elevation and computes per-point detection levels.
- Detection levels are calibrated using registration covariance and surface roughness.
- Semantic and instance segmentation refine geometry-based associations.
- The method targets city-scale 3D map change detection.
- It aims to overcome limitations of DSM and image differencing.
Entities
Institutions
- arXiv