三维点云数据骨架提取问题研究

发布时间：2018-04-17 11:56

本文选题：点云 + 骨架提取　；参考：《山东大学》2017年硕士论文

【摘要】：由于通过红外线等设备的扫描可以得到物体的坐标信息,所以随着计算机和扫描技术的进步,三维点云被越来越多的被应用在医学,动画,三维重建等研究领域。所以三维点云是目前比较流行的研究领域,而且此类问题也是研究的热点问题之一。骨架是一种可以简单的表达物体特征的简要结构。骨架有两种表现类型:一种是曲线模型,被称为曲线骨架。另一种为中轴面模型。需要应用的领域不同的话,也会拥有不同的需求。因此,就可以利用不同的需求对骨架进行不同的精简,精简完成之后的结构表示成为曲线骨架,因为这种骨架比较简单准确的表示了模型的拓扑信息同事又保留了模型的主要特征,所以这种形式的拓扑是比较常用的结构表示形式。骨架提取问题是比较热门的研究问题,通过对其进行研究,本文提出一种新的鲁棒的骨架提取方法。首先利用一种局部拉普拉斯方法对三维点云进行收缩,然后对收缩之后的部分点云进行主成分分析,提取局部点云骨架。对已提取骨架的局部点云进行固定,并将无法满足当前PCA半径条件的点云继续执行收缩。迭代执行收缩和局部骨架提取操作,直到所有点云满足给定的终止条件,并得到一个完整的点云骨架。完成上述过程之后的过程就是对初步得到骨架的优化和处理。本文中主要从以下这三个方面对骨架提取的问题进行改进:(1)本文基于拉普拉斯算子,提出了一种分步局部点云收缩方法,有效的防止局部点云数据过收缩,而其余部分点云数据收缩不够,无法提取其骨架的不均衡现象;(2)基于局部点云收缩的新方法,提出了一种基于动态PCA计算的骨架提取方法,在提取局部点云骨架时,能够动态调整PCA的计算半径,自动计算点云骨架分支,并最终将骨架分支合并起来;(3)优化骨架,对得到的骨架结果进行拓扑分类,并用三次B样条曲线进行拟合优化。实验结果表明,与其他已有方法相比,本文方法所得到的三维点云骨架结构质量更高且更加鲁棒。
[Abstract]:With the development of computer and scanning technology, 3D point cloud is more and more used in medicine, animation, 3D reconstruction and other research fields.Therefore, three-dimensional point cloud is a popular research field at present, and this kind of problem is also one of the hot issues.Skeleton is a simple structure that can express the features of objects.There are two types of skeleton: one is curve model, which is called curve skeleton.The other is the central axis model.Different areas of application will also have different requirements.As a result, you can use different requirements to streamline the skeleton differently, and the structure after the streamlining is represented as a curved skeleton.Because the skeleton can represent the topological information of the model simply and accurately, and retain the main characteristics of the model, the topology of this form is a more commonly used form of structural representation.Skeleton extraction is a hot research problem. By studying it, a new robust skeleton extraction method is proposed in this paper.Firstly, a local Laplace method is used to shrink the 3D point cloud, and then the local point cloud skeleton is extracted by principal component analysis (PCA).The local point cloud which has been extracted skeleton is fixed and the point cloud which can not satisfy the current PCA radius condition continues to shrink.The contractions and local skeleton extraction operations are performed iteratively until all point clouds satisfy the given termination conditions and a complete point cloud skeleton is obtained.The process of completing the above process is to optimize and deal with the initial skeleton.In this paper, we mainly improve the skeleton extraction problem from the following three aspects: 1) based on Laplace operator, we propose a step by step local point cloud shrinkage method, which can effectively prevent the local point cloud data from overshrinking.However, the other part of the point cloud data is not shrinking enough to extract its skeleton. (2) based on the new method of local point cloud shrinkage, a skeleton extraction method based on dynamic PCA computation is proposed, in which the local point cloud skeleton is extracted.The calculated radius of PCA can be adjusted dynamically, the point cloud skeleton branch can be calculated automatically, and finally the skeleton branch can be merged to form an optimized skeleton. The obtained skeleton results can be topologically classified and optimized by cubic B-spline curve fitting.The experimental results show that the proposed method has higher mass and robustness than other existing methods.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】