基于点云的植物骨架提取与建模研究

发布时间：2018-05-30 09:57

本文选题：植物建模 + Kinect　；参考：《江苏大学》2017年硕士论文

【摘要】：骨架作为三维模型的“紧凑”表示方法,能抽象反映出植物模型的拓扑结构以及体态特征。由于骨架的拓扑结构简单,操作方便,是三维模型表面重建、检索和匹配的基本要素,因而广泛应用于植物建模、三维动画设计、医学影像和测绘学图像等领域。现有的骨架提取算法多以人体和物体为初始模型,基于体素数据或网格曲面信息表示三维模型,而直接针对植物点云数据进行骨架提取方法却很少。由于植物形态结构复杂、自身遮挡比较严重以及常用采集设备的精度限制,采集到的植物初始云数据中存在大量的噪声,且局部数据缺失比较严重,如果采取传统的骨架提取方法,其正确性和完整性难以保证。因此,本文提出了一种基于Kinect的植物点云数据骨架提取的方法,具体内容为:(1)以Kd-tree作为数据结构进行点云的存储和处理,利用拾取技术从三维植物模型中交互选取所需点云数据,并用不同颜色作初始标记,对采集出的散乱点云进行去噪与配准处理。(2)通过搜索目标点的k近邻建立点云的拓扑关系,采用k均值聚类方法最大限度地保留植物枝干弯曲延伸特征的点云集合,并依据每个目标点与其相邻点的夹角关系生成植物骨架。(3)根据提取的植物骨架重建三维植物模型,并进行真实感渲染,与传统网格重建方法比较,本文方法避免了边缘失真、枝叶遮挡以及扫描漏洞等方面的问题,并较为真实地还原出植物模型的形态及细节。(4)结合OpenGL设计开发原型系统,实现骨架提取、表面重建及纹理贴图等功能,用户通过交互式界面控制参数,直观、生动地显示实验结果。
[Abstract]:As a "compact" representation of 3D models, skeleton can abstractly reflect the topological structure and body features of plant models. Because of its simple topological structure and convenient operation, skeleton is the basic element of 3D model surface reconstruction, retrieval and matching, so it is widely used in plant modeling, 3D animation design, medical image and surveying and mapping image and so on. Most of the existing skeleton extraction algorithms take human body and object as the initial model and represent the 3D model based on voxel data or mesh surface information. However, there are few methods to extract skeleton directly from plant point cloud data. Because of the complexity of plant morphology and structure, the serious occlusion and the precision limitation of common acquisition equipment, there is a lot of noise in the initial cloud data, and the lack of local data is serious. If the traditional skeleton extraction method is adopted, it is difficult to guarantee its correctness and completeness. Therefore, this paper proposes a method of extracting the skeleton of plant point cloud data based on Kinect. The concrete content is: 1) using Kd-tree as the data structure to store and process the point cloud. The point cloud data is interactively selected from the 3D plant model by pick-up technique, and different colors are used as initial markers to de-noise and register the collected scattered point cloud. The topological relationship of the point cloud is established by searching the k-nearest neighbor of the target point. The k-means clustering method is used to preserve the point cloud set of the plant branch bending and extending characteristics to the maximum extent, and according to the angle relationship between each target point and its adjacent points, the plant skeleton is generated. 3) based on the extracted plant skeleton, the three-dimensional plant model is reconstructed. Compared with the traditional mesh reconstruction method, this method avoids the problems of edge distortion, branch and leaf occlusion and scan loopholes, and so on. The prototype system is designed and developed by combining with OpenGL. The functions of skeleton extraction, surface reconstruction and texture mapping are realized. The user controls the parameters through interactive interface, which is intuitionistic. The results of the experiment are vividly displayed.
【学位授予单位】：江苏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】