A Semantic–Geometric Collaborative Framework for Roof Plane Segmentation in Airborne LiDAR Point Clouds

Roof plane segmentation from airborne light detection and ranging (LiDAR) point clouds provides planar patches for reconstructing 3D roof surfaces and their structural relationships. Existing methods still find it difficult to balance semantic awareness and geometric consistency, especially in the presence of ambiguous boundaries between adjacent roof planes, scattered outlier or non-roof points, and diverse roof structures. To address this problem, this paper proposes a semantic–geometric collaborative framework for roof plane segmentation that uses semantic segmentation results to guide geometric clustering. First, semantic segmentation identifies planar interiors and inter-plane boundaries, with an additional noise or non-roof class where applicable. Second, a k-nearest neighbor (KNN) local-priority clustering method is designed to generate the initial segmentation. It selects seeds from reliable planar interiors and uses a double-ended queue to guide roof plane clusters to grow from stable interior regions toward uncertain boundaries. Third, a multi-neighbor edge growth (MEG) strategy with adaptive residual constraints is introduced to refine the initial segmentation. For each initial roof plane, the plane model and residual threshold are estimated from its own geometric distribution, and the remaining unassigned non-noise points are assigned according to point-to-plane distances and neighboring plane relationships. Under the evaluation procedure used in this study, the proposed method achieves strong instance-level scores and competitive point-level scores on the WHU-Roof, RoofN3D, and Building3D datasets. On the WHU-Roof dataset, it achieves 86.8% coverage (Cov), 91.3% weighted coverage (WCov), and 97.4% F1-score. On the RoofN3D dataset, it achieves the highest Cov of 85.6%, with a WCov of 90.3%. On the Building3D dataset, it achieves 82.7% Cov, 88.4% WCov, and 95.8% F1-score. These results support combining semantic boundary information with geometric clustering to separate adjacent roof planes and refine their boundaries on the evaluated datasets.

Authors

Institutions

Publication Details

Journal
Remote Sensing
Published
2026-11-20
DOI
https://doi.org/10.3390/rs18224002
Primary Topic
Automated Road and Building Extraction
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Semantic–Geometric Collaborative Framework for Roof Plane Segmentation in Airborne LiDAR Point Clouds

Jingmin Tu, Li Li, 姚剑 Yao Jian, Jie Li et al.
4 links
Remote Sensing
Automated Road and Building Extraction
article

A Semantic–Geometric Collaborative Framework for Roof Plane Segmentation in Airborne LiDAR Point Clouds

Jingmin Tu, Li Li, 姚剑 Yao Jian, Jie Li, Wenjie Zhao, Deng Zhang
article en

Abstract

Roof plane segmentation from airborne light detection and ranging (LiDAR) point clouds provides planar patches for reconstructing 3D roof surfaces and their structural relationships. Existing methods still find it difficult to balance semantic awareness and geometric consistency, especially in the presence of ambiguous boundaries between adjacent roof planes, scattered outlier or non-roof points, and diverse roof structures. To address this problem, this paper proposes a semantic–geometric collaborative framework for roof plane segmentation that uses semantic segmentation results to guide geometric clustering. First, semantic segmentation identifies planar interiors and inter-plane boundaries, with an additional noise or non-roof class where applicable. Second, a k-nearest neighbor (KNN) local-priority clustering method is designed to generate the initial segmentation. It selects seeds from reliable planar interiors and uses a double-ended queue to guide roof plane clusters to grow from stable interior regions toward uncertain boundaries. Third, a multi-neighbor edge growth (MEG) strategy with adaptive residual constraints is introduced to refine the initial segmentation. For each initial roof plane, the plane model and residual threshold are estimated from its own geometric distribution, and the remaining unassigned non-noise points are assigned according to point-to-plane distances and neighboring plane relationships. Under the evaluation procedure used in this study, the proposed method achieves strong instance-level scores and competitive point-level scores on the WHU-Roof, RoofN3D, and Building3D datasets. On the WHU-Roof dataset, it achieves 86.8% coverage (Cov), 91.3% weighted coverage (WCov), and 97.4% F1-score. On the RoofN3D dataset, it achieves the highest Cov of 85.6%, with a WCov of 90.3%. On the Building3D dataset, it achieves 82.7% Cov, 88.4% WCov, and 95.8% F1-score. These results support combining semantic boundary information with geometric clustering to separate adjacent roof planes and refine their boundaries on the evaluated datasets.

Remote SensingVol. 18(22)
China University of Geosciences (CN), Wuhan University (CN), Hubei University of Technology (CN)
National Natural Science Foundation of China
Sustainable cities and communities
Automated Road and Building Extraction
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon