当前位置: X-MOL 学术Math. Probl. Eng. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
An Improved Building Reconstruction Algorithm Based on Manhattan World Assumption and Line-Restricted Hypothetical Plane Fitting
Mathematical Problems in Engineering ( IF 1.430 ) Pub Date : 2020-09-10 , DOI: 10.1155/2020/9267854
Xiaoguo Zhang 1 , Guo Wang 2 , Ye Gao 1 , Huiqing Wang 1 , Qing Wang 1
Affiliation  

An improved patch-based multiview stereo (PMVS) algorithm based on Manhattan world assumption and the line-restricted hypothetical plane fitting method according to buildings’ spatial characteristics is proposed. Different from the original PMVS algorithm, our approach generates seed points purely from 3D line segments instead of using those feature points. First, 3D line segments are extracted using the existing Line3D++ algorithm, and the 3D line segment clustering criterion of buildings is established based on Manhattan world assumption. Next, by using the normal direction obtained using the result of 3D line segment clustering, we propose a multihypothetical plane fitting algorithm based on the mean shift method. Then, through subdividing on the triangle mesh constructed based on the building hypothetical plane model, semidense point cloud can be quickly obtained, and it is used as seed points of the PMVS pipeline instead of the sparse and noisy seed points generated by PMVS itself. After that, dense point cloud can be obtained through the existing PMVS expansion pipeline. Finally, unit and integration experiments are designed; the test results show that the proposed algorithm is 15%∼23% faster than the original PMWS in running time, and at the same time, the reconstruction quality of buildings is improved as well by successfully removing many noise points in the buildings.

中文翻译:

基于曼哈顿世界假设和线受限假想平面拟合的改进建筑物重建算法

提出了一种基于曼哈顿世界假设的改进的基于补丁的多视点立体(PMVS)算法,并根据建筑物的空间特性,提出了行受限假想平面拟合方法。与原始PMVS算法不同,我们的方法仅从3D线段生成种子点,而不是使用这些特征点。首先,使用现有的Line3D ++算法提取3D线段,并基于Manhattan世界假设建立建筑物的3D线段聚类标准。接下来,通过使用通过3D线段聚类的结果获得的法线方向,我们提出了一种基于均值平移方法的多假设平面拟合算法。然后,通过细分基于建筑物假设平面模型构造的三角形网格,可以快速获得半密集点云,并将其用作PMVS管道的种子点,而不是PMVS本身生成的稀疏和嘈杂的种子点。之后,可以通过现有的PMVS扩展管道获得密集点云。最后,设计了单元和集成实验。测试结果表明,该算法在运行时间上比原始PMWS快15%〜23%,同时通过成功消除建筑物中的许多噪声点,提高了建筑物的重建质量。
更新日期:2020-09-10
down
wechat
bug