基于加权约束的单体建筑物点云表面重建算法
作者简介:王森援(1993-),女,山西长治人,硕士生,主要研究方向为测绘遥感技术和点云的三维重建算法。E-mail: collenwang@jmu.edu.cn
收稿日期: 2018-12-14
要求修回日期: 2019-01-20
网络出版日期: 2019-05-25
基金资助
国家自然科学基金项目(61702251)
福建省高校产学合作科技重大项目(2017H6015)
福建省自然科学基金项目(2016J01310、2016J01309)
厦门市科技局项目(3502Z20183032)
Algorithm based on Weighted Constraints for Reconstructing the Point Cloud Surface of Single-Building
Received date: 2018-12-14
Request revised date: 2019-01-20
Online published: 2019-05-25
Supported by
National Natural Science Foundation of China, No.61702251
Major Projects of Industry-University Cooperation in Fujian Province, No.2017H6015
Project supported by the Natural Science Foundation of Fujian Province, No.2016J01310, 2016J01309
Science and Technology Bureau of Xiamen, No.3502Z20183032.
Copyright
建筑物点云表面重建在高精度城市测绘、虚拟现实等领域有十分广泛的应用前景。由于建筑物的几何形态多变,重建算法普遍存在计算速率慢、拟合精度低和模型结构不完整的问题。为此,本文以单体建筑物为研究对象,提出基于加权约束的单体建筑物点云表面重建算法,在表面初始化过程中充分考虑数据对结构拟合的贡献。在此基础上,构建基于正则集的单体建筑物表面重建算法,实现建筑物拟合过程中的加权拟合误差、近邻结构平滑的同步优化。针对多类建筑物三维点云的实验结果表明,相比传统的建筑物重建策略,本文的加权约束方法可根据不同类型的点云数据设计自适应权重,并选择模型拟合中最优的权重函数,在高噪声、低精度点云数据下能得到更高精度的单体建筑物表面模型。
王森援 , 蔡国榕 , 王宗跃 , 吴云东 . 基于加权约束的单体建筑物点云表面重建算法[J]. 地球信息科学学报, 2019 , 21(5) : 654 -662 . DOI: 10.12082/dqxxkx.2019.180661
Building reconstruction based on 3D point cloud data has broad application prospects in fields such as high precision urban mapping and virtual reality. Due to the diverse geometry of buildings, there are widespread problems in traditional reconstruction algorithms, e.g., slow computation speed, low fitting precision, and incompleteness of building structures. Thus, with single-building as the research object, this paper proposed an algorithm based on weighted constraints for reconstructing point cloud surfaces. By fully considering each point’s contribution to the fitting plane during the surface initialization process, the proposed algorithm, which is based on regular sets, simultaneously optimizes the error of adaptively weighted fitting and the smoothness of neighbor structures. The algorithm was applied to the 3D point clouds of various buildings. Results showed that, compared with conventional building reconstruction strategies, the weighted-constraints based algorithm of this study can design adaptive weights according to different types of point clouds, and can choose the optimal weight for model fitting. In cases where the point cloud data contain high noise and low accuracy, the proposed algorithm can help generate more accurate surface models for single-building.
Fig. 1 Graphical illustration of χi→j图1 χi→j的图示表述 |
Fig. 2 Image of the weight functions of model fitting at 0-1图2 模型拟合的权重函数在0-1上的图像 |
Tab. 1 A regular-set point cloud surface reconstruction algorithm with weighted constraints表1 加权约束的正则集点云表面重建算法 |
输入:点云分割块{Sj}∈S,局部匹配候选集{Pi→i∈P},角度集θ 输出:正则集P*={Pi→j}, 模型之间的关系集{< Pi→j , Pi→k , rels >} |
---|
1:P0:={Pj→j}∈P // (1) 初始化 2:P*:=Ø // 候选集的初始集合 3:θ:=areamax/areaminρ // 估计面积阈值 4:while P*≠ Ø do // (2) 候选集生成 5: Pθ:={Pj→j,s.t.∀j area(Pj→j) > θ}∈P0 // 选择大于阈值的初始候选集 6: P0:=P0\Pθ 7: :=Pθ∪Enrich(Pθ , P*)∪Enrich(Pθ , Pθ ) // 扩充候选集 8: P*:=P*∪{P*⊆} //最小化能量方程 // (3) 选择正则集 if no break in enrich then decrease area threshold θ = θ / 2 // (4) 迭代 9:return P* 10:function Enrich(P, Pfixed , ||max=2300) 11:for each pair < Pj→j ∈ P , Pk→l ∈ Pfixed > do 12: if min r (|θr - ∠(ni,nk)|) < r 13: then :=∪{Pk→j} // 生成新的候选集 14: if || > ||max then break // 限制变量的数目 15:return // 候选集扩充结束 |
Tab. 2 Parameters of the experimental data表2 实验数据的参数 |
数据 | 输入点的个数 | 点密度/m |
---|---|---|
集装箱(Container) | 81 k | 0.08 |
楼房(Bungalow) | 71 k | 0.3 |
欧拉大楼(Euler) | 586 k | 0.004 |
组合箱(Boxunion) | 100 k | 0.02 |
帝国大厦(Empire) | 1.2 M | 0.0025 |
噪声的组合箱(Boxunion_noise) | 100 k | 0.08 |
Fig. 3 Experimental data图3 实验数据 |
Tab. 3 Comparison of result errors using weight functions of model fitting表3 模型拟合的权重函数的结果误差对比 |
方法 | 均方根误差/m | 分配给形状的点的比例/% |
---|---|---|
CnvPCA2 CncPCA LinePCA CnvPCA4 Classic LDPCA BDPCA | 7.975 14.072 14.065 12.165 15.039 14.002 14.334 | 72.83 26.97 26.98 49.33 13.19 26.94 26.01 |
Fig. 4 Visual reconstruction effect of container图4 集装箱的可视化重建效果 |
Fig. 5 Visual reconstruction effect of Bungalow图5 楼房的可视化重建效果 |
Tab. 4 Comparison of result errors using weight functions of model fitting表4 模型拟合的权重函数的结果误差对比 |
方法 | 均方根误差/m | 分配给形状的点的比例/% |
---|---|---|
CnvPCA2 CncPCA LinePCA CnvPCA4 Classic LDPCA BDPCA | 0.0614 0.0654 0.0632 0.0647 0.0689 0.0703 0.0627 | 77.614 77.019 76.437 77.612 77.014 76.462 77.612 |
Fig. 6 Visual reconstruction effect of Euler图6 欧拉大楼的可视化重建效果 |
Fig. 7 Visual reconstruction effect of Boxunion图7 组合箱的可视化重建效果 |
Tab. 5 Comparison of result errors using weight functions of model fitting表5 模型拟合的权重函数的结果误差对比 |
方法 | 均方根误差/楼高/m | 分配给形状的点的比例/% |
---|---|---|
CnvPCA2 CncPCA LinePCA CnvPCA4 Classic LDPCA BDPCA | 0.491 0.499 0.476 0.496 1.724 10.570 0.466 | 80.75 80.50 80.64 80.68 80.50 78.62 80.64 |
Fig. 8 Visual reconstruction effect of Empire tower图8 帝国大厦塔身的可视化重建效果 |
Fig. 9 Visual reconstruction effect of the tower top图9 帝国大厦塔顶的可视化重建效果 |
Fig. 10 Visual reconstruction effect of Boxunion_noise图10 含噪声的组合箱的可视化重建效果 |
The authors have declared that no competing interests exist.
[1] |
|
[2] |
|
[3] |
|
[4] |
|
[5] |
|
[6] |
|
[7] |
|
[8] |
[
|
[9] |
[
|
[10] |
|
[11] |
|
[12] |
|
[13] |
|
[14] |
|
[15] |
|
[16] |
|
[17] |
|
[18] |
|
[19] |
|
[20] |
|
[21] |
|
[22] |
|
[23] |
|
[24] |
|
[25] |
|
/
〈 | 〉 |