基于BSP树的点云精简方法研究

【作者】 赵强；

【导师】 彭国华；

【作者基本信息】 西北工业大学 ， 计算数学， 2007， 硕士

【摘要】 逆向工程指的是根据现有的模型，利用数字化测量设备获取实体数据，然后对这些数据进行拟合，构建一个完整的实体模型。逆向工程的流程一般包括数据获取，数据预处理，曲面拟合和实体模型生成四个部分。由于目前采用的数据获取设备为激光测量设备，它可以从产品表面获取大量的测量数据，可达到几十万或几百万。如此巨大的数据量，不仅加大了系统的负荷，而且大大降低后续处理的效率，所以必须事先进行数据精简处理。 本文主要研究基于点的简化方法，首先介绍如何根据空间数据点的位置关系计算曲面的几何信息，如单位法向量和曲率等。在法向调整问题上，本文借鉴武剑洁的“环推衍法”思想，提出一种新的法向调整方法。 其次，对Pauly等人提出的层级聚类方法进行讨论，主要工作是对此方法的不足之处进行修正： 1．原方法把计算局部点云的曲面变化最大方向的方法，用来对全局点云进行曲面最大变化方向的计算，没有考虑此局部计算方法对全局点云的适用性。本文考虑到距中心不同点对计算曲面变化最大方向的不同影响，采用加权系数对该影响进行修正。 2．原方法中选取聚类单元中心点为代表点，因为中心点不一定是原始采样点，会造成重建的曲面上出现非预期的凸出和凹入三角片。本文以距中心最近点为代表点，可以有效避免了重建误差产生的可能性。 3．相近表而情形存在，将会导致重建模型出现非流形错误。本文根据Jong和Lee判断相近表面方法，提出一种修正相近表面的方法，该方法包含两部分：对于包含三点以上的聚类单元，引入辅助点判断相近表面；对于含三点及以下的聚类单元进行强制细分避免相近表面。 4．采样不足情形即当聚类单元满足共平面特性时，单元中包含点的数目较多，若采用一个代表点会造成局部区域的“空白”，从而使重建产生非预期空洞。本文根据实验结果提出一种限制标准：相邻树节点层数不能相差太多。对不满足该限制树节点继续向下切分，产生辅助代表点，从而避免采样不足。 最后，对不同的实体模型的点云数据进行实验，结果证明了改进后的方法具有一定的实用性。更多还原

【Abstract】 The definition of Reverse Engineering (RE) is that given a three dimensional real model, constructing an integrated Computer Aided Designed model which can be. The digital equipment can be used to acquire the discrete data also called point data. Therefore, RE is divided into four processes: data acquisition, data preprocessing, surface fitting and model construction. The data is so large that is not only burdens the system, but also reduces the efficiency of the later processes. So it is very important to simplify the original data.This paper mainly studies the simplification methods. At first, this paper introduces how to estimate the surface geometric characteristics, such as curvature and normal vector, according to analyze the space position relation among the point cloud data. In this part, a new method based on Wu’s work is proposed to adjust the direction of the normal vectors.The work in this paper is that improving the hierarchical clustering algorithm on the way to choose the partition plane and the representative sample, and to avoid the problem that near surface and under-sampling.1. Using the method, which is applicable to local surface, to computing the maximum variation of the surface, some error may occur. Considering the different points having different distances to the center, this can bring some influences. This paper uses weighting coefficient to remove the influences. 2. If the representative sample is the center of a cell, then because it is not the original point, some unexpected triangles may protrude from and concave into surface. This study chooses the original point that is the closest to the center, and selecting the closest point can effectively reduce the probability of producing errors.3. The near surface can cause non-manifold. The paper is based on Jong’s and Lee’s algorithm, and proposes a method to modify the near surface problem. The method contains two parts: auxiliary point is used to detect the near surface in the cell containing more than three points, and the cell containing less than three points is subdivided compulsively.4 Unexpected holes during reconstruction resulting from under-sampling, due to insufficient information about for neighboring points in the local region. To更多还原