徽派建筑构件点云的曲面重构研究

发布时间：2018-05-02 14:42

本文选题：点云 + 预处理　；参考：《安徽建筑大学》2017年硕士论文

【摘要】：曲面重构技术在逆向工程、机器视觉、虚拟现实VR(Virtual Reality)、现代医疗等多个领域中被广泛使用。随着三维激光扫描技术的飞速发展和日益成熟,可通过三维激光扫描技术来获取古建筑物表面高密度的三维点云数据信息,再通过后期的分割优化、滤波去噪、法线计算、曲面重构等操作完成对点云数据的三维建模。由于支持曲面重构的软件多存在操作复杂、交互定义过多、对输入数据要求过于严苛等问题至今未得到广泛推广。然而,对古建筑进行三维建模的通常做法是前期通过人工测量得到建筑物中各个构件的具体物理参数,后期再借助三维建模软件等方式进行描绘、贴图等,最终效果不够真实且误差较大。通过对徽派建筑中大小构件进行考察,本课题最终选用了徽派建筑所特有的穿斗抬梁式木构架、门楼和栏杆作为研究对象从而展开研究。论文围绕徽派建筑构件点云的曲面重构系统研究主要探讨了以下三方面问题:第一,点云数据预处理工作。论文以徽派建筑所特有的穿斗抬梁式木构架为例,对其进行多站点扫描获得原始点云数据。首先,通过改进的迭代最近点算法(简称ICP算法,Iterative Closet Point)对多站点云数据进行配准拼接,获得整栋建筑物点云数据;其次,对配准后点云数据使用随机采样一致性算法进行分割优化,获得梁柱点云数据;最后,利用体素网格化法对梁柱点云数据中离群点进行采样,再通过滤波器Statistical OutlierRemoval对离群点进行剔除操作。第二,基于预处理后点云数据的曲面重构。通过使用最小二乘法对预处理后点云数据进行直接地法线估计推断,得到梁柱点云数据中每一点的法线及其正负向。借助点云库PCL(Point Cloud Library)开源平台,分别使用贪婪投影三角化算法、BallPivoting算法以及泊松算法对经过预处理且具有法线的梁柱点云数据进行曲面重构并同时给出了核心算法步骤,最终得到每种算法相对应的效果图及参数信息。第三,曲面重构结果对比。通过对比三种算法在同一构件点云条件下的曲面重构效果图以及网格面数、网格顶点数、孔洞数、耗时等参数信息,结果表明泊松算法最终效果图要明显优于其它两种方法,尤其是最终结果中没有孔洞出现,这对于曲面重构最终三维模型的完整性输出至关重要。以上研究借助三维激光扫描技术获取被测对象的点云原始数据,通过多边形网格化等技术实现点云数据的分割、滤波、法线估计等预处理操作,最终建立基于泊松算法的徽派建筑构件点云数据曲面重构模型。文中所提曲面重构整套研究方案为三维建模技术在建筑中的应用尤其是在古建筑复建、数字化等领域中提供了新的研究思路。
[Abstract]:Surface reconstruction is widely used in many fields such as reverse engineering, machine vision, virtual reality VR(Virtual reality, modern medicine and so on. With the rapid development and maturity of 3D laser scanning technology, 3D laser scanning technology can be used to obtain high density 3D point cloud data information on the surface of ancient buildings, and then through the later segmentation optimization, filtering and denoising, normal calculation, Surface reconstruction and other operations complete the 3D modeling of point cloud data. Due to the complexity of operation, the definition of interaction and the strict requirement of input data, the software supporting surface reconstruction has not been widely popularized up to now. However, the common method of 3D modeling of ancient buildings is to obtain the physical parameters of each component by manual measurement in the early stage, and to depict them in the later stage by means of 3D modeling software, such as mapping, mapping, and so on. The final effect is not real and the error is large. Through the investigation of the large and small components in the Huizhou architecture, this topic finally selects the unique wooden frame of the bucket lift beam, the gatehouse and the railing as the research object to carry out the research. This paper mainly discusses the following three problems about the surface reconstruction system of Huizhou architectural component point cloud: first, point cloud data preprocessing. Taking Huizhou architecture as an example, the original point cloud data is obtained by multi-site scanning. Firstly, the improved iterative nearest point algorithm (ICP algorithm) is applied to the registration of multi-site cloud data to obtain the whole building point cloud data. After registration point cloud data is segmented and optimized by random sampling consistency algorithm to obtain Liang Zhu point cloud data. Finally, we use voxel mesh method to sample outliers in Liang Zhu point cloud data. Then the outliers are removed by filter Statistical OutlierRemoval. Second, surface reconstruction based on pre-processing point cloud data. By using the least square method to infer directly the normals of pre-processed point cloud data, the normals and their positive and negative directions of each point in Liang Zhu's point cloud data are obtained. With the help of point cloud library PCL(Point Cloud library open source platform, the greedy projection triangulation algorithm BallPivoting algorithm and Poisson algorithm are used to reconstruct the surface of preprocessed and normal Liang Zhu point cloud data, and the core algorithm steps are given. Finally, the corresponding effect diagram and parameter information of each algorithm are obtained. Third, the surface reconstruction results are compared. By comparing the surface reconstruction results of the three algorithms under the same component point cloud condition, as well as the grid surface number, the number of mesh vertices, the number of holes, the time consuming and so on, the results show that the Poisson algorithm is better than the other two methods in the final effect graph. In particular, there are no holes in the final results, which is very important for the integrity output of the final 3D model reconstruction. The above research uses 3D laser scanning technology to obtain the original point cloud data of the measured object, and realizes the preprocessing operations such as point cloud data segmentation, filtering, normality estimation and so on through polygonal gridding technology. Finally, the point cloud data surface reconstruction model based on Poisson algorithm is established. The whole research scheme of surface reconstruction in this paper provides a new research idea for the application of 3D modeling technology in architecture, especially in the field of reconstruction and digitization of ancient buildings.
【学位授予单位】：安徽建筑大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TU198;P225.2

【参考文献】