摘要
法向信息是点云表面重建的关键属性,而屏蔽泊松重建算法在法向不准确区域未进行针对性约束,导致容易出现过拟合问题。为此提出了一种基于法向约束的泊松算法,通过计算点云的初始和最终法向之间的偏离程度确定点云法向估计的准确度,以此来进行针对性约束。一方面将法向准确度与屏蔽泊松方程中采样点权重联系起来,在每个分辨率上以获取更为准确的位置约束;另一方面在利用多重网格算法进行自适应求解时,对法向准确度较低的点所在的八叉树节点不进行扩展,以减少伪曲面的生成和内存开销。在多种不同特点的标准数据集上得到的对比实验结果表明,本文算法能更好地平衡过度平滑和过拟合的问题,在整体上提高了重建曲面的准确度,同时也在一定程度上减少了运行时间和内存开销。
Normal information is the key attribute for point cloud surface reconstruction,however the screened Poisson reconstruction algorithm does not perform targeted constraints in areas with normal inaccuracies,resulting in overfitting problem.This paper proposes a Poisson algorithm based on normal constraints.The accuracy of point cloud normal estimation is determined by calculating the degree of deviation between the initial and final,which performs targeted constraints.On the one hand,the normal accuracy is related to the weight of the sampling points in the screened Poisson equation to obtain more accurate position constraints at each resolution;on the other hand,when using the multi-grid algorithm for adaptive solution,the octree nodes where the normal accuracy is lower are not expanded to reduce pseudo-surfaces and memory overhead.The results obtained on several datasets show that our algorithm can better balance the problems of over-smoothing and over-fitting,improve the precision of the surface and computational efficiency.
作者
鲁猛胜
姚剑
董赛云
LU Mengsheng;YAO Jian;DONG Saiyun(School of Remote Sensing and Information Engineering,Wuhan University,Wuhan 430079,China)
出处
《测绘地理信息》
CSCD
2022年第4期51-55,共5页
Journal of Geomatics
基金
国家重点研发计划(2017YFB1302400)
关键词
三维点云
泊松重建
法向约束
八叉树
3D point cloud
poisson reconstruction
normal constraint
octree
作者简介
第一作者:鲁猛胜,硕士,主要从事点云处理和三维重建方法研究。E-mail:lumengsheng@whu.edu.cn;通讯作者:姚剑,教授,主要从事计算机视觉、三维重建方法研究。E-mail:jian.yao@whu.edu.cn
