逆向工程点云数据预处理技术研究

发布时间：2018-05-12 19:41

本文选题：三维点云数据 + K近邻搜索　；参考：《哈尔滨理工大学》2017年硕士论文

【摘要】：随着三维点云数据采集技术的不断发展,实体表面的数字化变得简单,同时精度更高。但是无论采用什么样的数据采集方法都会受到人为、环境等因素的影响,不可避免的存在测量误差,这些误差的存在将直接影响到三维重构模型的质量。同时随着测量精度的提高,采集到的物体表面数据点的数量十分的庞大,影响三维重构的效率。本文主要研究逆向工程中三维散乱点云数据处理技术,主要研究内容如下:在栅格空间划分的基础上,以给定的栅格边长对包围盒内数据点进行第一次空间划分,计算出划分后“黑体”占有率。在第一次栅格划分的基础上,综合考虑点云总数、点云的分布情况以及黑体“黑体”占有率,并定义了有效黑体,使得对于不同特征的点云数据其栅格边长的划分更加合理,加快K近邻搜索的速度。点云数据中的离群点和噪声点影响三维点云数据重构的质量,对这些噪声滤除同时保持物体表面的几何特征,提出了一种基于法矢修正的点云数据去噪平滑算法。采用数理统计方法滤除点云数据中的离群点,对于噪声点通过加权协方差矩阵估算初始法向量和曲率,将具有相似几何特征的点限制在法向量相似的区域,进而在三边滤波因子修正法矢的基础上,三边平滑降噪点的位置。该算法不仅能够对大的噪声信号有较好的滤除效果,对小噪声平滑的同时保持点云数据模型的尖锐及边缘特征。对常用的数据精简算法进行分析,提出了一种将曲率和法向量结合的数据精简算法,综合考虑曲率和法向量方向两个因素,避免了点云数据的过度精简。
[Abstract]:With the development of 3D point cloud data acquisition technology, the digitization of solid surface becomes simpler and more accurate. However, no matter what kind of data acquisition method is adopted, it will be affected by man-made, environmental and other factors, and there will inevitably be measurement errors, which will directly affect the quality of 3D reconstruction model. At the same time, with the improvement of measurement accuracy, the number of collected surface data points is very large, which affects the efficiency of 3D reconstruction. In this paper, the data processing technology of 3D scattered point cloud in reverse engineering is studied. The main contents are as follows: on the basis of grid space partition, the data points in the bounding box are partitioned for the first time with the given grid side length. Calculate the share of the "black type" after the division. On the basis of the first grid partition, the total number of point clouds, the distribution of point clouds and the "share of blackbody" are considered synthetically, and the effective blacktype is defined, which makes the partition of grid length for point cloud data with different characteristics more reasonable. Speed up K nearest neighbor search. The outliers and noise points in point cloud data affect the reconstruction quality of 3D point cloud data. A smoothing algorithm for point cloud data denoising based on normal vector correction is proposed to remove these noises and preserve the geometric features of the object surface at the same time. A mathematical statistical method is used to filter outliers in point cloud data. For noise points, the initial normal vectors and curvature are estimated by weighted covariance matrix, and the points with similar geometric characteristics are restricted to areas with similar normal vectors. On the basis of triangular-filtering factor modifying normal vector, triangulation smoothes the position of denoising point. The algorithm can not only filter the large noise signals but also keep the sharp and edge features of the point cloud data model while smoothing the small noise. Based on the analysis of common data reduction algorithms, a data reduction algorithm combining curvature and normal vector is proposed, which considers two factors, curvature and normal vector direction, and avoids the excessive reduction of point cloud data.
【学位授予单位】：哈尔滨理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP274

【相似文献】