基于深度霍夫投票的建筑点云轻量级表面重建

doi:10.3785/j.issn.1008-973X.2026.02.012

浙江大学学报(工学版)

2026, Vol. 60

Issue (2): 341-350 DOI: 10.3785/j.issn.1008-973X.2026.02.012

计算机技术与控制工程

基于深度霍夫投票的建筑点云轻量级表面重建

陈佳舟1(

),朱肖航1,徐阳辉1,高崟2,3,鲁一慧4,毛真4,李胜龙4,章超权2

1. 浙江工业大学计算机科学与技术学院，浙江杭州 310023
2. 莫干山地信实验室，浙江德清 313200
3. 国家基础地理信息中心，北京 100830
4. 山东省国土测绘院，山东济南 250102

Lightweight surface reconstruction method for building point clouds based on deep Hough voting

Jiazhou CHEN1(

),Xiaohang ZHU1,Yanghui XU1,Yin GAO2,3,Yihui LU4,Zhen MAO4,Shenglong LI4,Chaoquan ZHANG2

1. College of Computer Science and Technology, Zhejiang University of Technology, Hangzhou 310023, China
2. Moganshan Geospatial Information Laboratory, Deqing 313200, China
3. National Geomatics Center of China, Beijing 100830, China
4. Shandong Provincial Institute of Land Surveying and Mapping, Jinan 250102, China

全文: PDF(1954 KB) HTML

摘要：

针对实景三维场景中建筑物结构缺失、数据冗余、噪声多等问题，提出新的建筑点云轻量级表面重建方法，进行建筑的多边形网格模型重建. 构建高效的建筑数据集生成框架，自动生成包含5 500个带标签的建筑模型数据. 针对建筑点云中平面提取困难的问题，使用深度霍夫投票预测建筑平面，采用基于面的非极大值抑制算法(F-NMS)有效去除预测的重复面以及错误面. 设计建筑平面相邻关系预测模块，对经过非极大值抑制后的建筑平面进行相邻关系的预测. 定量实验结果表明，与如PolyFit的传统方法相比，所提方法在拟合精度与场景适应性方面均具有显著优势. 使用所提方法重建的建筑多边形网格模型保留了输入建筑点云的主要结构特征，存储量不到原始点云的1%.

关键词： 三维点云; 建筑简化; 三维重建; 霍夫投票; 网格模型

Abstract:

To address missing structures, data redundancy and noise in real-world 3D scenes, a lightweight surface reconstruction method for building point clouds was proposed that reconstructs polygonal mesh models. An efficient framework for building-dataset generation was proposed, automatically producing 5 500 labeled building models. To ease the plane extraction for building point clouds, the building planes were predicted using deep Hough voting, and a face-based non-maximal suppression algorithm (F-NMS) was used to efficiently remove the predicted duplicate and erroneous surfaces. A building plane adjacency prediction module was designed to predict the adjacency of the building planes after the F-NMS. Quantitative experimental results demonstrate that, compared to traditional methods such as PolyFit, the proposed approach exhibits significant advantages in both fitting accuracy and scene adaptability. The polygonal mesh models reconstructed by the proposed method retain the main structural features of the input building point clouds, with storage requirements reduced to less than 1% of the original point cloud data.

Key words: 3D point cloud building simplification 3D reconstruction Hough voting mesh model

收稿日期: 2025-07-17 出版日期: 2026-02-03

CLC:

TP 391.41

基金资助: 国家自然科学基金资助项目（62172367）；浙江省尖兵领雁计划研发攻关计划项目（2025C01073）.

作者简介: 陈佳舟（1984—），男，副教授，从事计算机图形学、人工智能研究. orcid.org/0000-0003-2780-6146. E-mail：cjz@zjut.edu.cn

	服务
	把本文推荐给朋友
	加入引用管理器
	E-mail Alert
	作者相关文章
	陈佳舟
	朱肖航
	徐阳辉
	高崟
	鲁一慧
	毛真
	李胜龙
	章超权

引用本文:

陈佳舟,朱肖航,徐阳辉,高崟,鲁一慧,毛真,李胜龙,章超权. 基于深度霍夫投票的建筑点云轻量级表面重建[J]. 浙江大学学报(工学版), 2026, 60(2): 341-350.

Jiazhou CHEN,Xiaohang ZHU,Yanghui XU,Yin GAO,Yihui LU,Zhen MAO,Shenglong LI,Chaoquan ZHANG. Lightweight surface reconstruction method for building point clouds based on deep Hough voting. Journal of ZheJiang University (Engineering Science), 2026, 60(2): 341-350.

链接本文:

https://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2026.02.012 或 https://www.zjujournals.com/eng/CN/Y2026/V60/I2/341

图 1 多边形网格重建的算法流程

图 2 基于深度霍夫投票的平面预测网络结构图

图 3 平面非极大值抑制算法的输入输出

图 4 配对面注意力模块

图 5 基于平面及相邻关系的建筑线框模型转换过程

图 6 建筑多边形网格模型

图 7 建筑点云重建结果及过程示例

表 1 投票点损失函数不同权重的消融实验（σ=0.01）

表 2 配对面注意力模块消融实验（σ=0.01）

表 3 基于面的非极大值抑制算法的相似度阈值消融实验（σ=0.02）

表 4 建筑简化性能对比评估表（σ=0.02）

图 8 三维建筑轻量化重建的不同方法可视化对比(σ=0.02)

图 9 建筑简化抗噪性可视化对比

表 5 建筑简化抗噪性定量评估结果

1	LUO H, ZHANG J, LIU X, et al Large-scale 3D reconstruction from multi-view imagery: a comprehensive review[J]. Remote Sensing, 2024, 16 (5): 773 doi: 10.3390/rs16050773
2	于海洋, 封顺天, 崔立鹏面向城市数字孪生的多尺度三维建模方法研究[J]. 电子技术应用, 2022, 48 (7): 78- 80 YU Haiyang, FENG Shuntian, CUI Lipeng Research on multi-scale 3D modeling method for urban digital twin[J]. Application of Electronic Technique, 2022, 48 (7): 78- 80
3	SCHNABEL R, WAHL R, KLEIN R Efficient RANSAC for point-cloud shape detection[J]. Computer Graphics Forum, 2007, 26 (2): 214- 226 doi: 10.1111/j.1467-8659.2007.01016.x
4	RABBANI T, VAN DEN HEUVEL F A, VOSSELMANN G Segmentation of point clouds using smoothness constraint[J]. International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences, 2006, 36 (5): 248- 253
5	QI C R, YI L, SU H, et al. PointNet++: deep hierarchical feature learning on point sets in a metric space [EB/OL]. (2017−06−07)[2024−03−14]. https://arxiv.org/pdf/1706.02413.
6	WANG R, HUANG S, YANG H. Building3D: an urban-scale dataset and benchmarks for learning roof structures from point clouds [C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Paris: IEEE, 2024: 20019–20029.
7	YANG G, XUE F, ZHANG Q, et al. UrbanBIS: a large-scale benchmark for fine-grained urban building instance segmentation [C]// Proceedings of the ACM SIGGRAPH 2023 Conference Proceedings. Los Angeles: ACM, 2023: 1–11.
8	PETERS R, DUKAI B, VITALIS S, et al Automated 3D reconstruction of LoD2 and LoD1 models for all 10 million buildings of the Netherlands[J]. Photogrammetric Engineering and Remote Sensing, 2022, 88 (3): 165- 170 doi: 10.14358/PERS.21-00032R2
9	NAN L, WONKA P. PolyFit: polygonal surface reconstruction from point clouds [C]// Proceedings of the IEEE International Conference on Computer Vision. Venice: IEEE, 2017: 2372–2380.
10	BOUZAS V, LEDOUX H, NAN L Structure-aware building mesh polygonization[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2020, 167: 432- 442 doi: 10.1016/j.isprsjprs.2020.07.010
11	HUANG J, STOTER J, PETERS R, et al City3D: large-scale building reconstruction from airborne LiDAR point clouds[J]. Remote Sensing, 2022, 14 (9): 2254 doi: 10.3390/rs14092254
12	BAUCHET J P, LAFARGE F City reconstruction from airborne lidar: a computational geometry approach[J]. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2019, 4: 19- 26
13	CHEN Z, SHI Y, XIONG Z, et al. Polyhedron-based graph neural network for compact building model reconstruction [C]// Proceedings of 2023 IEEE International Geoscience and Remote Sensing Symposium. Pasadena: IEEE, 2023: 923–926.
14	CHEN Z, SHI Y, NAN L, et al PolyGNN: polyhedron-based graph neural network for 3D building reconstruction from point clouds[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2024, 218: 693- 706 doi: 10.1016/j.isprsjprs.2024.09.031
15	HE X, LV C, HUANG P, et al. WindPoly: polygonal mesh reconstruction via winding numbers [C]// Computer Vision – ECCV 2024. [S.l.]: Springer, 2024: 294–311.
16	HUANG S, WANG R, GUO B, et al. PBWR: parametric-building-wireframe reconstruction from aerial LiDAR point clouds [C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2024: 27778–27787.
17	LI L, SONG N, SUN F, et al Point2Roof: end-to-end 3D building roof modeling from airborne LiDAR point clouds[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2022, 193: 17- 28 doi: 10.1016/j.isprsjprs.2022.08.027
18	JAVEED M A, GHAFFAR M A, ASHRAF M A, et al Lane line detection and object scene segmentation using otsu thresholding and the fast Hough transform for intelligent vehicles in complex road conditions[J]. Electronics, 2023, 12 (5): 1079 doi: 10.3390/electronics12051079
19	SYED M H, KUMAR S. Road lane line detection based on ROI using Hough transform algorithm [C]// Proceedings of Third International Conference on Computing, Communications, and Cyber-Security. Singapore: Springer, 2023: 567–580.
20	MATARNEH S, ELGHAISH F, AL-GHRAIBAH A, et al An automatic image processing based on Hough transform algorithm for pavement crack detection and classification[J]. Smart and Sustainable Built Environment, 2025, 14 (1): 1- 22 doi: 10.1108/SASBE-01-2023-0004
21	QI C R, LITANY O, HE K, et al. Deep Hough voting for 3D object detection in point clouds [C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Seoul: IEEE, 2019: 9276–9285.
22	ZHANG J Q, DUAN H B, CHEN J L, et al HoughLaneNet: lane detection with deep Hough transform and dynamic convolution[J]. Computers and Graphics, 2023, 116: 82- 92
23	ZHAO H, JIANG L, JIA J, et al. Point transformer [C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Montreal: IEEE, 2022: 16239–16248.
24	LI J, ZHOU J, XIONG Y, et al. An adjustable farthest point sampling method for approximately-sorted point cloud data [C]// Proceedings of the IEEE Workshop on Signal Processing Systems. Rennes: IEEE, 2022: 1–6.
25	HU J, SHEN L, SUN G. Squeeze-and-excitation networks [C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018: 7132–7141.
26	Republic of Estonia Land and Spatial Development Board. Building 3D model data [EB/OL]. [2024−04−15]. https://geoportaal.maaamet.ee/eng/Download-3D-data-p837.html.
27	DONG L, XIAO Y, LI Y, et al A collision detection algorithm based on sphere and EBB mixed hierarchical bounding boxes[J]. IEEE Access, 2024, 12: 62719- 62729
28	WU M, SUN M, ZHANG F, et al A fault detection method of electric vehicle battery through Hausdorff distance and modified Z-score for real-world data[J]. Journal of Energy Storage, 2023, 60: 106561

[1]	何婧瑶,李鹏飞,汪承志,吕振鸣,牟萍. 基于双目视觉和改进YOLOv8的动态三维重建方法[J]. 浙江大学学报(工学版), 2025, 59(7): 1443-1450.
[2]	邢志伟,朱书杰,李彪. 基于改进图卷积神经网络的航空行李特征感知[J]. 浙江大学学报(工学版), 2024, 58(5): 941-950.
[3]	兰欢,余建波. 基于深度学习三维成型的钢板表面缺陷检测[J]. 浙江大学学报(工学版), 2023, 57(3): 466-476.
[4]	陈涵娟,达飞鹏,盖绍彦. 基于竞争注意力融合的深度三维点云分类网络[J]. 浙江大学学报(工学版), 2021, 55(12): 2342-2351.
[5]	陈文卓, 陈雁, 张伟明, 何少炜, 黎波, 姜俊泽. 圆弧面动态空气喷涂数值模拟[J]. 浙江大学学报(工学版), 2018, 52(12): 2406-2413.
[6]	王盛,项志宇. 基于多谱融合的植被环境中障碍物检测[J]. 浙江大学学报(工学版), 2015, 49(11): 2223-2229.
[7]	杨力, 刘俊毅, 王延长, 刘济林. 基于全景相机和全向激光雷达的致密三维重建[J]. 浙江大学学报(工学版), 2014, 48(8): 1481-1487.
[8]	赖小波,朱世强,方纯洁. 一种复杂背景图像三维重建算法及其医学应用[J]. J4, 2012, 46(11): 2061-2067.
[9]	王炜强, 徐进, 杜歆, 刘济林. 基于宽基线立体视觉的远距离三维重建[J]. J4, 2010, 44(6): 1073-1078.
[10]	李根潘翔陈志杨张三元叶修梓. 基于泊松形状信号的网格模型部件分割算法[J]. J4, 2008, 42(12): 2074-2079.
[11]	张晓洁刘旭刘向东孟晓. 旋转LED阵列体显示系统的仿真平台构建[J]. J4, 2005, 39(10): 1541-1544.

Viewed

Full text

Abstract

Cited

Shared

Discussed