摘要
点云的区域分割实质上是根据点的局部几何特性的相似性对点进行分类 ,利用自组织特征映射神经网络(self- organizing feature m ap,SOFM)可以实现无监督的特征聚类 ;本文使用 SOFM进行反向工程中点云的区域分割 ,选用数据点的坐标、法向量六维向量作为 SOFM的输入 ,通过改进 SOFM的学习算法 ,加入输入权和距离权 ,加速了分割的速度和正确性。利用 SOFM方法实现点云分割 ,具有以下优点 :不必限定面的类型 ;用户可以控制分区的个数 ;可以处理噪音数据。并用实际数据验证了此方法的可行性。
Segmentation of point-cloud aims at classifying point-cloud into several subspaces and each subspace can be fitted to a surface. In this paper, segmentation using self-organizing feature map (SOFM) network is used for such segmentation. Six dimensional feature vector (3-dimensional coordinate and 3-dimensional normal vector) was taken as input to SOFM. Weighted input and weighted Euclidean distance were adopted in learning process of SOFM, which improved the speed and accuracy of segmentation. Segmentation by SOFM is robust to noise, and has no limitation for surface type. Finally, the method is validated by real scanned point-cloud.
出处
《机械科学与技术》
CSCD
北大核心
2002年第4期659-661,共3页
Mechanical Science and Technology for Aerospace Engineering
关键词
自组织特征映射
神经网络
数据分割
反向工程
Self-organizing feature map
Neural networks
Point-cloud segmentation
