点云数据特征提取算法的改进

发布时间：2018-04-21 02:35

本文选题：三维激光扫描 + 点云　；参考：《东华理工大学》2013年硕士论文

【摘要】：点云特征的提取在人脸识别，计算机视觉，数字娱乐，地形学等领域有非常重要的使用价值。而三维激光扫描的诞生，给予点云数据全新的生命，如何对点云数据进行特征解剖，是现在刻不容缓的技术要求。传统的特征提取技术并不能满足现在技术要求，如何改进算法来获得更加完美的点云各要素特征，是本文主要的研究内容。针对激光扫描获得的散乱数据，本文一方面介绍传统的点、线、面特征提取方法，另一方面研究如何对特征进行快速、准确的提取。主要研究内容有： 1）叙述了贝塞尔、B-样条和NURBS三种方法在曲线和曲面提取中的原理，并用Matlab分别作图比较了三种方法提取线、面的优缺点； 2）对基于曲率极值法及其改进法提取特征点进行阐述并进行实例比较分析； 3）叙述折线生长法，采用设定阀值减少候选特征点来改进特征线的提取，利用手掌数据进行实例研究分析； 4）分析传统最小二乘的缺点，并对其加以改进，，采用高程分域，把点云数据分块，然后对分块数据进行最小二乘，相临边界用最小二乘值平均值拟合，最后加以光滑处理。采用Matlab对16个控制点进行拟合比较；利用了2组不同类型的数据整体对其进行特征点、线、面的提取，对本文改进的算法进行分析验证。从实验例证可以得出：本文算法对表面复杂目标的点、线提取更加有优势；在面提取分块上有待进一步简化，但整体连贯性好。
[Abstract]:Point cloud feature extraction is very important in face recognition, computer vision, digital entertainment, topography and other fields. The birth of 3D laser scanning gives a new life to point cloud data. How to dissect the point cloud data is an urgent technical requirement. The traditional feature extraction technology can not meet the current technical requirements. How to improve the algorithm to obtain more perfect features of the elements of the point cloud is the main research content of this paper. Based on the scattered data obtained by laser scanning, this paper introduces the traditional methods of feature extraction of points, lines and surfaces on the one hand, and studies how to extract features quickly and accurately on the other hand. The main research contents are as follows: 1) the principle of Besselger B-spline and NURBS in curve and surface extraction is described, and the advantages and disadvantages of the three methods are compared by Matlab. 2) the feature points are extracted based on curvature extremum method and its improved method, and the examples are compared and analyzed. 3) the method of broken line growth is described. The method of reducing candidate feature points by setting threshold value is used to improve the extraction of feature lines, and the palm data is used for case study and analysis. 4) the shortcomings of traditional least squares are analyzed and improved. The point cloud data is partitioned by elevation domain, then the block data is partitioned by least square method, and the boundary is fitted with the mean value of least square value, and finally the smooth processing is carried out. The 16 control points were fitted and compared by Matlab. Two groups of different types of data are used to extract the feature points, lines and surfaces, and the improved algorithm is analyzed and verified. From the experimental examples, it can be concluded that the algorithm has more advantages in line extraction for points with complex surface targets, and needs to be further simplified in the area of surface extraction, but the overall coherence is good.
【学位授予单位】：东华理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P225.2

【引证文献】