尖锐特征曲面点云模型各向异性邻域搜索  

THE ANISOTROPIC NEIGHBORHOOD SEARCHING FOR POINT CLOUD MODEL WITH SHARP FEATURE SURFACE

作　　者：袁小翠[1] 陈华伟 Yuan Xiaocui;Chen Huawei(Jiangxi Key Laboratory of Precision Drive and Control,Nanchang Institute of Techno log y,Nanchang 330099,Jiangxi,China;School of Mechanical and Electrical Engineering,Guizhou Normal University,Guiyang 550025,Guizhou,China)

机构地区：[1]南昌工程学院江西精密驱动与控制重点实验室,江西南昌330099 [2]贵州师范大学机械与电气工程学院,贵州贵阳550025

出　　处：《计算机应用与软件》2018年第12期253-258,共6页Computer Applications and Software

基　　金：江西省教育厅科学技术研究项目(GJJ61122);国家自然科学基金项目(51365037)

摘　　要：特征曲面点云模型的特征点k邻域各向同性。基于此邻域对点云数据处理容易造成严重的处理误差,因此提出一种各向异性邻域搜索方法。对点云建立KD树搜索k邻域,采用邻域拟合法估计点云法向量和检测特征点。将特征点的各向同性邻域映射到高斯球形成不同的数据簇,层次聚类法对高斯球上的数据分类,高斯球上最大类对应的邻域为最优各向异性子邻域。为了验证算法的有效性,将该方法与KD树和栅格邻域搜索方法进行应用和耗时比较。实验结果表明,该方法耗时最长,但是应用于点云数据处理,如点云法向量估算和点云去噪时效果更佳。The k neighborhood of feature points of feature surface cloud model is isotropic. Based on this neighborhood, it is easy to cause serious processing errors in point cloud data processing, so an anisotropic neighborhood searching was proposed. KD tree was used to search k neighborhood for point cloud, and neighborhood fitting method was used to estimate point cloud normal vector and detect feature points. The isotropic neighbors of feature points were mapped into Gaussian sphere to form different data-clusters, and the hierarchical clustering method was used to classify the data in Gaussian sphere. The neighborhood corresponding to the largest class on Gauss sphere was the optimal anisotropic sub neighborhood. To verify the validity of the method, the proposed method was compared with KD tree and grid neighborhood searching in terms of application and time consumption. Experimental results show that the proposed method is the most time-consuming, but it is more effective when applied to point cloud data processing, such as point cloud normal vector estimation and point cloud denoising.

关 键 词：点云邻域 高斯映射 法向量估计 层次聚类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]