【作者】 金龙存；

【导师】 万旺根；

【作者基本信息】 上海大学 ， 通信与信息系统， 2012， 博士

【摘要】 本文工作目的是对3D点云复杂曲面重构中的关键算法进行理论研究。激光扫描仪可采集得到具有复杂表面结构模型的原始点云数据。本文研究了原始点云数据的预处理问题;实现了基于点云的空洞插值算法;研究了变分法隐式曲面理论;针对复杂的点云表面实现了快速变化表面重构算法;并对较大点云模型数据实现了点云模型动态载入算法。取得了以下创新成果:首先,针对原始点云数据中存在的噪声,提出了一种新的点云抗噪算法,能有效去除点云噪声,并对点云噪声有一定的抑制作用,通过对比实验验证了算法性能;第二,根据实际问题中模型点云数据的结构特征,一般模型点云数据存在着空洞问题,研究利用Hermite曲面理论和细节描述子算法,提出了一种新的空洞插值算法,能有效对点云表面中存在的空洞进行恢复,实验结果说明了空洞插值算法的有效性;第三,分析了点云表面理论模型,针对表面细节恢复上的难题,提出了变分法隐式曲面重构算法,可较好地重构模型细节部分,对比实验结果验证了算法的优越性能;第四,结合空洞插值算法和变分法隐式曲面算法,提出了一种新的快速表面重构算法,通过对比实验说明了算法具有优良的性能指标,可有效重构3D点云模型的快速变化表面;最后,根据实际情况,针对大规模点云数据从外存无法一次性地载入内存的问题,提出了点云模型动态载入算法,利用有效的降维算法进行了点云模型的精简,并实现了模型的实时重构。通过研究五个关键算法,经对比实验表明,文中提出的算法具有良好的性能,为实现3D点云复杂曲面重构及后续应用提供了良好的途径。更多还原

【Abstract】 The purpose of the work in this thesis is to tackle key schemes for complex sur-face reconstruction from point clouds. A 3D laser scanner can get original point clouds from model of complex surface. This thesis tackles pre-processing of original point clouds. It tackles hole interpolation method from point clouds. The thesis ad-dresses variational implicit surface scheme and tackles reconstructing surface from complex surface of point clouds. And then, it proposes dynamic loading method for large point cloud data of a model. The main contribution of this work is followed:Firstly, this thesis proposes a novel method for anti-noise of point clouds. The method can de-noise and be robust to noise. Experimental results show it is high level performance. Secondly, according to reality problem of configuration in point clouds, it proposes a novel scheme for filling holes in a point cloud. The scheme uti-lizes Hermite surface theory and algorithm of depictor in detail for data configura-tion. Experimental results show the proposed scheme is efficient. Thirdly, it tackles efficient mathematic model from point cloud. The thesis proposes a novel method for variational implicit surface reconstruction from complex surface of a point cloud. The method can reconstruct surface of detail. Comparative experimental results ve-rify high performance of the proposed method. Fourthly, the thesis proposes a noval scheme for a variety of surface reconstruction. The scheme utilizes novel schemes of filling holes and variational implicit surface reconstruction. Experimental results show the scheme is efficient for a variety of 3D surface reconstruction. Finally, ac-cording to reality problem of massive point cloud, it proposes dynamic method for a model of point cloud loading. It also utilizes reduction algorithm to reduce models of point cloud. It tackles surface real-time reconstructing a model.The thesis tackles five key algorithms. Experimental results show the proposed algorithms are of high performance. Providing an efficient access, the four key algo-rithms implement complex surface reconstruction from 3D point cloud and a new solution for applications.更多还原