论文总字数：26454字

摘 要

随着虚拟现实技术的快速发展，对自然景观的仿真模拟获得了越来越多研究者的关注。树木是自然景观不可或缺的一部分，它们形态各异，结构复杂，生存在多变的自然环境中。因此真实树的重建比较困难，它对自然景观的模拟起着至关重要的作用。本文以重建出自然环境中的真实树木为目标，具体展开了以下几项工作：

（1）研究并实现了算法提取并匹配树木特征点。实验表明,该算法适用于树木这一特殊对象，对38张梧桐树图片进行特征点提取与匹配只需要40秒；

（2）实现了基于法和光束平差法的树木稀疏重建。通过匹配点的信息计算得到基础矩阵以及本质矩阵，并由此求解出投影矩阵，利用三角关系还原空间点的位置，得到稀疏的树木点云模型。梧桐树重建实验表明稀疏重建只需要41秒，但得到的点云数目较少；

（3）实现了基于法的树木稠密重建。利用法加密稀疏重建后的点云，获得树木更加细节的特征。然后将匹配点构建成面片，通过对面片进行扩展并去除其中误差较大的面片最终实现稠密重建。实验表明稠密重建得到的点云数据更多，效果更好，但重建的时间为22分钟，远大于稀疏重建；

（4）实现了基于点的树木骨架提取。将点云模型分割后获取各区域的点并光顺连成粗骨架，通过细化和下采样获得最终的树木拓扑结构。对比实验表明该算法对于点云数据不完整的情况也能获得清晰的骨架。

关键字：树木，特征点匹配，三维重建，点云数据，骨架提取

ABSTRACT

With the rapid development of virtual reality technology, the simulation of natural landscape has received more and more researchers' attention. Trees are an integral part of the natural landscape, they are of different shapes, complex structures, living in the changing natural environment. Therefore, it is challenging to reconstruct of real trees, which plays an important role in the simulation of natural landscape. The goal of this article is to rebuild real trees in natural environment, specifically to carry out the following work:

(1) The SIFT algorithm for feature point extraction and matching is studied and realized, which is suitable for the special object of trees. Experiment showed that it took only 40 seconds to extract and match the feature points of 38 pictures of sycamore tree.

(2) The sparse reconstruction of trees based on RANSAC method and beam leveling method is realized. The basic matrix and the essential matrix are calculated through the information of the matching point, and the projection matrix is solved. The position of the spatial point is reduced by the triangular relation, and the sparse tree point cloud model is obtained. The reconstruction of the sycamore trees shows that sparse reconstruction only takes 41 seconds, but the number of point clouds is small;

(3) Realization of dense reconstruction of trees based on PMVS method. Using PMVS method to encrypt sparse reconstructed point cloud, get more detailed characteristics of trees. Then the matching points are constructed into dough pieces, which then are expanded and filtered to realize the dense reconstruction. Experiments show that the dense reconstruction acquired more point cloud data and the effect is better, but the reconstruction time of 22 minutes, much larger than the sparse reconstruction;

(4) The tree point cloud skeleton is extracted through ROSA points in this paper. First dividing the point cloud, calculating the ROSA points of each region to obtain the coarse skeleton of the tree. After that, connecting and refining it to acquire the final tree skeleton. Experiments showed that the algorithm is also applicable to the case where the point cloud data is incomplete, but a clear topology can still be obtained.

KEY WORDS: tree, feature point matching, 3D reconstruction, point cloud, skeleton extraction

目 录

摘 要 I

ABSTRACT II

第一章 绪 论 1

1.1 研究背景 1

1.2 国内外的研究现状 1

1.3 本文研究内容和论文结构安排 2

第二章 树木特征点提取及匹配算法验证及选择 4

2.1 引言 4

2.2 特征点提取算法分析 4

2.2.1 Moravec算子 4

2.2.2 Harris算法 5

2.2.3 分析总结 5

2.3 树木二维特征点提取算法 5

2.4 树木二维特征点匹配算法 7

2.4.1 特征点匹配方法 7

2.4.2 误匹配点删除方法 7

2.5 实验结果与分析 8

2.6 本章小结 9

第三章 树木三维重建算法研究 10

3.1 引言 10

3.2 稀疏3D重建 10

3.2.1 基础矩阵与本质矩阵 10

3.2.2 两幅图像三维点坐标求解 11

3.2.3 多幅图像投影矩阵求解 12

3.2.4 光束法平差 13

3.3 稠密3D重建 13

3.3.1 基本概念 14

3.3.2 PMVS重建过程 15

3.4 实验结果与分析 15

3.5 本章小结 17

第四章 基于ROSA点的树木点云骨架提取算法 18

4.1 引言 18

4.2 骨架提取算法要求 18

4.2.1 拓扑特性 18

4.2.2 度量特性 19

4.2.3 可靠性 19

4.3 骨架提取过程 20

4.3.1 输入模型预处理 20

4.3.2 点云分割 21

4.3.3 粗骨架提取 22

4.3.4 曲线骨架提取 24

4.4 实验结果与分析 26

4.5 本章小结 28

第五章 总结 29

5.1 工作总结 29

5.2 展望 29

参考文献 30

致 谢 32

剩余内容已隐藏，请支付后下载全文，论文总字数：26454字