激光点云的混合流形谱聚类自适应分割方法

发布时间：2018-03-26 00:50

本文选题：成像系统　切入点：三维成像　出处：《光学学报》2017年10期

【摘要】：将激光点云视为分布于三维欧氏空间的线性与非线性混合流形,提出一种基于混合流形谱聚类的自适应点云分割方法。由混合概率主成分分析法构造的M个主成分分析器组成混合概率模型,得到描述点云的邻接矩阵;将点云分割的几何特征在谱空间进行降维嵌入,利用N-cut方法得到描述点云分割特征的多维向量;结合类间类内划分算法自适应分割点云。实验结果表明,对于三种受测点云,所提出的算法能在较宽预设参数范围内以80%以上概率得到收敛于几何特征的分割结果,参数稳定性较好。在对点云添加均值为0,标准差为0.01的高斯噪声与0.25倍数量的离群点复合噪声的情况下,算法表现出良好的抗噪性;将该算法应用于切片式激光三维成像的卫星模型点云中也取得了理想分割结果。
[Abstract]:The laser point cloud is regarded as a linear and nonlinear hybrid manifold distributed in three dimensional Euclidean space. An adaptive point cloud segmentation method based on hybrid manifold spectral clustering is proposed. The mixed probability model is composed of M principal component analyzers constructed by mixed probabilistic principal component analysis and the adjacent matrix describing the point cloud is obtained. The geometric feature of point cloud segmentation is embedded in spectral space, and the multi-dimensional vector describing the feature of point cloud segmentation is obtained by using N-cut method, and the point cloud is segmented adaptively by using intra-class partition algorithm. The experimental results show that, for the three measured point clouds, The proposed algorithm can obtain the segmentation results converging to geometric features by more than 80% probability in a wide range of preset parameters. The parameter stability is good. When Gao Si noise with mean value 0, standard deviation 0.01 and outliers compound noise of 0.25 times are added to the point cloud, the algorithm shows good anti-noise performance. The algorithm is applied to the satellite model point cloud in slice laser 3D imaging and the ideal segmentation result is obtained.
【作者单位】：中国人民解放军装备学院研究生管理大队;中国人民解放军装备学院光电装备系;
【分类号】：TN249;TP391.41

【相似文献】