一种融合多特征聚类集成的室内点云分割方法

摘要

针对特定场景下传统点云分割算法不精确及特征描述不全面的问题,提出一种融合2D和3D多特征的近邻传播(AP)聚类集成分割方法.从点云中获得一组表征复杂室内场景不同点云类别的描述子,如彩色图像特征、曲率、法向量、旋转图像等,根据它们之间的差异性,通过对每类特征进行AP聚类得到聚类成员,建立聚类成员簇间一致性矩阵,并利用Ncut算法进行图分割获得最终的点云分割结果.实验结果表明,该算法相较传统的点云分割算法能更准确地区分室内复杂三维点云场景,并且具有更好的稳定性.%Aiming at the problem that the traditional point cloud segmentation algorithm is not precise and feature description is not comprehensive in specific scenes,an Affinity Propagation (AP) clustering ensemble segmentation method fusing with 2D and 3D features is proposed.Firstly,a set of descriptors representing different cloud types of complex indoor scenes,such as colour image features,curvature,normal vectors,rotating images,are obtained from point clouds.Secondly,according to the difference between them,the clustering members are obtained by AP clustering for each class of features,and the cluster consensus matrix is established.Finally,the final segmentation result is obtained by using Ncut algorithm.Experimental results show that the proposed method is better than traditional point cloud segmentation algorithm in distinguishing indoor 3D point cloud scene,and has better stability.