A semantic-assisted method for constructing multi-floor indoor spatial topology models from 3D point clouds

doi:10.11947/j.AGCS.2026.20250382

Abstract

Abstract:

The indoor space topological model serves as the fundamental data basis for precise navigation and personalized location services. However, the hierarchical complexity and flexible topological characteristics of large-scale indoor public spaces pose limitations for existing methods in multi-floor space partitioning and topological relationship representation. To address this, this paper proposes a semantic-assisted method for constructing multi-floor indoor spatial topology models from 3D point clouds. First, semantic information from 3D point clouds is leveraged to assist traditional height histogram models and watershed segmentation, achieving an initial multi-level “floor-room” partitioning of indoor subspaces. Next, the local connectivity between adjacent subspaces is quantitatively described using semantic constraints such as walls, floors, and ceilings. A 3D Markov random field (MRF) model is then employed to optimize the preliminary over-segmented subspaces, resulting in locally continuous and globally consistent multi-floor indoor space partitioning. Finally, doors and staircases are used to reconstruct intra-floor and inter-floor subspace connectivity, respectively, and a complete 3D indoor topological model is established based on the IndoorGML standard. Experimental results demonstrate that the proposed method achieves recall, mean intersection over union (mIoU), and precision of 97.87%, 96.05%, and 98.09% for floor segmentation, and 89.87%, 85.66%, and 94.81% for room segmentation, with a normalized graph edit distance (nGED) of 81.35% for topological relationship representation. Compared with existing state-of-the-art methods, the proposed approach improves the average precision and recall of space partitioning by approximately 9.60% and 9.30%, respectively, and enhances nGED by over 2.49%, demonstrating greater stability and reliability.

Key words: 3D point clouds, spatial subdivision, indoor topological modeling, 3D Markov random model, IndoorGML

CLC Number:

P237

Zhengwen WANG, Juntao YANG, Zhizhong KANG, Yutao ZHANG, Xuzhe WANG, Xue ZHANG. A semantic-assisted method for constructing multi-floor indoor spatial topology models from 3D point clouds[J]. Acta Geodaetica et Cartographica Sinica, 2026, 55(5): 909-926.

Add to citation manager EndNote|Reference Manager|ProCite|BibTeX|RefWorks

URL: http://xb.chinasmp.com/EN/10.11947/j.AGCS.2026.20250382

http://xb.chinasmp.com/EN/Y2026/V55/I5/909

Figures/Tables 32

Fig. 1

Fig. 2

Fig. 3

Fig. 4

Fig. 5

Fig. 6

Fig. 7

Fig. 8

Fig. 9

Tab. 1

Fig. 10

Fig. 11

Fig. 12

Fig. 13

Tab. 2

Tab. 3

Tab. 4

Fig. 14

Fig. 15

Fig. 16

Tab. 5

Tab. 6

Tab. 7

Fig. 17

Fig. 18

Fig. 19

Tab. 8

Fig. 20

Fig. 21

Fig. 22

Fig. 23

Tab. 9

References 23

[1]	KANG Zhizhong, YANG Juntao, YANG Zhou, et al. A review of techniques for 3D reconstruction of indoor environments[J]. ISPRS International Journal of Geo-Information, 2020, 9(5): 330.
[2]	HUGHES N, CHANG Yun, HU Siyi, et al. Foundations of spatial perception for robotics: hierarchical representations and real-time systems[J]. The International Journal of Robotics Research, 2024, 43(10): 1457-1505.
[3]	ISIKDAG U, ZLATANOVA S, UNDERWOOD J. A BIM-oriented model for supporting indoor navigation requirements[J]. Computers, Environment and Urban Systems, 2013, 41: 112-123.
[4]	闫金金, 尚建嘎, 余芳文, 等. 面向实时定位的室内空间结构分析及制图方法[J]. 武汉大学学报(信息科学版), 2016, 41(8): 1079-1086.
	YAN Jinjin, SHANG Jianga, YU Fangwen, et al. Indoor spatial structure and mapping methods for real-time localization[J]. Geomatics and Information Science of Wuhan University, 2016, 41(8): 1079-1086.
[5]	郭慧霞, 韩李涛, 张鹏飞. 基于点云的室内房间要素及拓扑关系提取[J]. 北京测绘, 2023, 37(2): 160-166.
	GUO Huixia, HAN Litao, ZHANG Pengfei. Extracting room elements and topological relations from indoor point cloud[J]. Beijing Surveying and Mapping, 2023, 37(2): 160-166.
[6]	史云飞, 刘克辉, 李向伟, 等. 一种含室内空间结构的建筑物3D LOD概念模型与生成方法[J]. 武汉大学学报(信息科学版), 2022, 47(4): 561-569, 596.
	SHI Yunfei, LIU Kehui, LI Xiangwei, et al. A 3D LOD conceptual model of building with indoor spatial structure and its generating method[J]. Geomatics and Information Science of Wuhan University, 2022, 47(4): 561-569, 596.
[7]	ARMENI I, SENER O, ZAMIR A R, et al. 3D semantic parsing of large-scale indoor spaces[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 1534-1543.
[8]	李琪琪, 花向红, 赵不钒, 等. 一种室内场景点云平面分割的新方法[J]. 中国激光, 2021, 48(16): 1604002.
	LI Qiqi, HUA Xianghong, ZHAO Bufan, et al. New method for plane segmentation of indoor scene point cloud[J]. Chinese Journal of Lasers, 2021, 48(16): 1604002.
[9]	QI C R, YI Li, SU Hao, et al. PointNet++: deep hierarchical feature learning on point sets in a metric space[EB/OL]. [2025-12-10]. https://arxiv.org/abs/1706.02413.
[10]	双丰, 黄兴文, 李勇, 等. 基于深度学习的大规模点云语义分割方法综述[J]. 测绘科学, 2023, 48(2): 195-209.
	SHUANG Feng, HUANG Xingwen, LI Yong, et al. A survey of large-scale point cloud semantic segmentation based on deep learning[J]. Science of Surveying and Mapping, 2023, 48(2): 195-209.
[11]	LI Feng, SHI Wenzhong, ZHANG Hua, et al. Automatic multi-storey building partitioning from point cloud via bidirectional cloth simulation[J]. Journal of Building Engineering, 2024, 97: 110854.
[12]	LIU Yudong, HUANG Han, GAO Ge, et al. Dataset and benchmark for as-built BIM reconstruction from real-world point cloud[J]. Automation in Construction, 2025, 173: 106096.
[13]	沈乐, 李桂清, 冼楚华, 等. 室内3D点云模型的门窗检测[J]. 计算机辅助设计与图形学学报, 2019, 31(9): 1494-1501.
	SHEN Le, LI Guiqing, XIAN Chuhua, et al. Door and window detection in 3D point cloud of indoor scenes[J]. Journal of Computer-Aided Design & Computer Graphics, 2019, 31(9): 1494-1501.
[14]	OCHMANN S, VOCK R, KLEIN R. Automatic reconstruction of fully volumetric 3D building models from oriented point clouds[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2019, 151: 251-262.
[15]	NIKOOHEMAT S, DIAKITÉ A A, ZLATANOVA S, et al. Indoor 3D reconstruction from point clouds for optimal routing in complex buildings to support disaster management[J]. Automation in Construction, 2020, 113: 103109.
[16]	YANG Juntao, KANG Zhizhong, ZENG Liping, et al. Semantics-guided reconstruction of indoor navigation elements from 3D colorized points[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2021, 173: 238-261.
[17]	GAO Xiang, YANG Ronghao, CHEN Xuewen, et al. A new framework for generating indoor 3D digital models from point clouds[J]. Remote Sensing, 2024, 16(18): 3462.
[18]	KELLNER M, STAHL B, REITERER A. Reconstructing geometrical models of indoor environments based on point clouds[J]. Remote Sensing, 2023, 15(18): 4421.
[19]	WANG Y, JIAO W. Parameter-driven level of detail derivation method for semantic building facade model[J]. Journal of Geodesy & Geoinformation Science, 2024, 7(3): 57-75.
[20]	BOYKOV Y, VEKSLER O, ZABIH R. Fast approximate energy minimization via graph cuts[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2001, 23(11): 1222-1239.
[21]	CHANG A, DAI A, FUNKHOUSER T, et al. Matterport3D: learning from RGB-D data in indoor environments[EB/OL]. [2025-12-10]. https://arxiv.org/abs/1709.06158.
[22]	HUGHES N, CHANG Yun, CARLONE L. Hydra: a real-time spatial perception system for 3D scene graph construction and optimization[EB/OL]. [2025-12-10]. https://arxiv.org/abs/2201.13360.
[23]	WERBY A, HUANG Chenguang, BÜCHNER M, et al. Hierarchical open-vocabulary 3D scene graphs for language-grounded robot navigation[C]//Proceedings of the 1st Workshop on Vision-Language Models for Navigation and Manipulation at ICRA. [S.l.]: IEEE, 2024.

房间	墙线	门数	房间类型
R0	（E₀，E₃，E₈，E₁₀，E₁₆，E₁₄，E₇，E₂）	3	走廊
R₁	（E₁，E₄，E₉，E₃）	1	普通房间
R₂	（E₅，E₇，E₁₂，E₁₁，E₆）	1	普通房间
R₃	（E₁₂，E₁₄，E₁₅，E₁₃）	1	普通房间

场景	点云数量	面积/m²	房间数	窗户数	门数
区域1	4.402 6×10⁷	24.7×48.3	41	31	47
区域2	4.731 5×10⁷	30.9×51.3	40	8	53
区域3	1.866 2×10⁷	29.3×25.7	23	8	22
区域4	4.347 0×10⁷	47.3×26.6	49	31	59
区域5	7.864 9×10⁷	66.4×45.0	68	51	75
区域6	4.135 3×10⁷	23.2×45.4	48	31	50

场景	点云数量	面积/m²	房间数	窗户数	门数
7y3	6.199×10⁶	22.45×14.40	36	37	41
D7n	8.918×10⁶	27.18×35.10	43	36	44
px4	9.321×10⁶	22.91×41.89	49	65	45
q9v	4.897×10⁶	27.46×18.29	22	11	30
e9z	4.553×10⁶	17.54×14.87	30	28	22
zsn	3.764×10⁶	22.41×13.45	20	22	15
pa4	8.800×10⁶	27.84×30.30	47	47	45
Jmb	4.126×10⁶	16.08×12.39	22	28	15
aay	4.003×10⁶	21.87×11.13	18	17	22
QUC	7.692×10⁶	29.09×44.74	37	59	53
5Lp	5.245×10⁶	23.98×19.48	30	12	19
qoi	4.471×10⁶	12.58×25.28	23	24	29
VFu	6.704×10⁶	31.04×35.82	35	17	49

场景	楼层	语义辅助高度直方图结果			MRF优化结果
场景	楼层	召回率	交并比	精确率	召回率	交并比	精确率
e9z	1层	66.52	66.521	00.00	96.08	96.08	100.00
	2层	99.92	92.87	92.93	99.93	98.52	98.59
	3层	99.04	98.96	99.92	99.04	98.96	99.94
Jmb	1层	90.78	90.32	99.45	99.22	98.67	99.45
Jmb	2层	99.31	88.19	88.73	99.38	98.52	99.13
pa4	1层	99.93	98.36	98.43	99.93	98.36	98.43
	2层	97.52	92.42	94.65	97.13	96.15	98.95
	3层	95.75	92.48	96.44	98.99	97.53	98.52
	4层	81.41	80.81	99.10	96.74	95.02	98.17
qoi	1层	99.91	94.51	94.59	99.86	98.57	98.70
qoi	2层	95.50	95.43	99.92	98.88	98.76	99.88
VFu	1层	99.92	98.93	99.01	99.92	98.93	99.01
	2层	99.62	93.46	93.79	99.39	97.79	98.38
	3层	91.01	91.01	100.00	97.51	97.13	99.61
平均值		96.95	94.57	97.51	97.87	96.05	98.09

场景	楼层	语义辅助直方图结果			MRF优化结果
场景	楼层	召回率	交并比	精确率	召回率	交并比	精确率
e9z	1层	81.82	81.82	100.00	90.91	90.91	100.00
	2层	72.73	72.73	100.00	90.91	90.91	100.00
	3层	100.00	100.00	100.00	100.00	100.00	100.00
Jmb	1层	75.00	69.23	90.00	91.67	91.67	100.00
Jmb	2层	70.00	63.64	87.50	90.00	90.00	100.00
pa4	1层	66.67	66.67	100.00	83.33	83.33	100.00
	2层	73.68	63.64	82.35	89.47	80.95	89.47
	3层	77.78	70.00	87.50	88.89	88.89	100.00
	4层	100.00	80.00	80.00	100.00	100.00	100.00
qoi	1层	90.00	90.00	100.00	90.00	90.00	100.00
qoi	2层	92.31	92.31	100.00	84.62	84.62	100.00
VFu	1层	81.82	69.23	81.82	81.82	75.00	90.00
	2层	80.00	75.00	92.31	84.62	84.62	100.00
	3层	88.89	80.00	88.89	88.89	80.00	88.89
平均值		87.85	83.02	93.45	89.87	85.66	94.81