局部形状特征概率混合的半自动三维点云分类

发布时间：2018-04-27 18:01

本文选题：点云分类 + 局部形状推断　；参考：《浙江大学学报(理学版)》2017年01期

【摘要】：三维激光扫描获取的点云数据可用于数字城市建设、三维模型获取、场景分析与物体测量等领域.但因遮挡和噪声的影响,加之扫描场景复杂,采样精度受限,使得不能直接运用经典的曲面和三维空间理论对点云数据进行有效分析和处理.分类是点云数据预处理的重要方式之一.提取近邻四面体体积、近邻法向量差异度、主方向差异度和主曲率值4个局部形状特征,采用概率混合策略构建了一种点云数据的半自动分类方法,可实现平面点集、柱面点集和其他点集的有效区分.其中,概率混合策略是依据近邻点平均距离和单指标类别一致程度估计每个特征推断形状的概率,通过混合加权,依据概率赋权函数最大值准则进行局部形状推断.可实现用户交互,以便处理不同扫描尺度和精度的点云数据.采用本文方法对模拟生成的点云、单棵树木点云、街道场景点云、旷野自然场景扫描点云以及航空机载扫描点云等多组数据进行了实验,结果表明,基于局部形状特征的概率混合方法对各种点云数据均具有良好的分类效果.
[Abstract]:The point cloud data obtained by 3D laser scanning can be used in the fields of digital city construction, 3D model acquisition, scene analysis and object measurement. However, due to the influence of occlusion and noise, combined with the complexity of scanning scene and limited sampling accuracy, the classical surface and three-dimensional space theory can not be directly used to effectively analyze and process point cloud data. Classification is one of the important methods of point cloud data preprocessing. Four local shape features, namely, nearest neighbor tetrahedron volume, nearest neighbor normal vector difference, principal direction difference and principal curvature value, are extracted. A semi-automatic classification method for point cloud data is constructed by using probability mixing strategy, which can realize the planar point set. The efficient distinction between cylindrical point sets and other point sets. The probability mixing strategy is to estimate the probability of each feature inferring shape according to the average distance of the nearest neighbor point and the consistency degree of the single index class, and to infer the local shape according to the maximum value criterion of the probability weighting function by mixing the weights. User interaction can be implemented to process point cloud data with different scanning scales and accuracy. In this paper, the simulated data of point cloud, single tree point cloud, street scene point cloud, field natural scene scanning point cloud and airborne scanning point cloud are tested. The results show that, The probabilistic mixing method based on local shape features has a good classification effect on all kinds of point cloud data.
【作者单位】：北京林业大学理学院;中国科学院自动化研究所模式识别国家重点实验室;
【基金】：国家自然科学基金资助项目(61372190,61372168,61572502,61571439) 国家863计划课题项目(2015AA016402)
【分类号】：TP391.41

【相似文献】