基于角点稳健提取的长方体参数化拟合方法  

Parametric fitting algorithm for cuboid based on robust corner extraction

作　　者：马程远 沙建军 王艳恒 韩明 谭聪 Ma Chengyuan;Sha Jianjun;Wang Yanheng;Han Ming;Tan Cong(College of Intelligent Systems Science and Engineering,Harbin Engineering University,Harbin 150001,China;Qingdao Innovation Development Base,Harbin Engineering University,Qingdao 266000,China)

机构地区：[1]哈尔滨工程大学智能科学与工程学院,哈尔滨150001 [2]哈尔滨工程大学青岛创新发展基地,青岛266000

出　　处：《仪器仪表学报》2023年第3期271-279,共9页Chinese Journal of Scientific Instrument

基　　金：国防科技基础加强项目(KYY200541-2);装备预先研究项目(202-YJGFGC-C01-F003)资助。

摘　　要：长方体基元拟合是三维点云几何拟合的典型问题,在三维重建、逆向工程、工业三维量测等实际场景中有着广泛应用。在实际应用中由于遮挡及设备盲区等原因,通常无法获取完整长方体点云数据,导致建模或量测时难以准确拟合长方体结构。针对该问题,本文提出一种结合平面拟合投影分割以及残缺平面垂角检测的长方体参数化拟合方法实现了长方体参数化拟合。首先,该方法通过平面拟合投影分割算法获取强轮廓信息的平面点云;然后,设计了残缺平面垂角检测算法,以拟合长方体真实角点;最后,利用非共面四点法对长方体残缺角点进行计算补全,获得完整的长方体参数信息。实验表明,本文方法在各类情况下均能准确检测并补全长方体角点信息以及平面参数信息,角点查准率召回率均为100%,平均误差仅为1.204×10-3m,能够实现精确的长方体参数化拟合。The cuboid primitive fitting is a typical problem of 3D point cloud geometric fitting,which is widely used in practical scenes,such as 3D reconstruction,reverse engineering and industrial 3D measurement.In practical applications,it is usually impossible to obtain complete cuboid point cloud data due to occlusion and equipment blind area,which makes it difficult to accurately fit cuboid structure in modeling or measurement.To solve this problem,this article proposes a cuboid parameterized fitting method which combines plane-fitting projection segmentation and incomplete plane vertical angle detection,and realizes the cuboid parameterized fitting.Firstly,the plane point cloud with strong contour information is obtained by the plane fitting projection segmentation algorithm.Then,the detection algorithm of incomplete plane vertical angle is designed to fit the real corner points of cuboids.Finally,the non-coplanar fourpoint method is used to calculate and complete the incomplete corner points of cuboids to obtain the complete cuboid parameter information.Experiments show that the proposed method can accurately detect and complete cuboid corner point information and plane parameter information in all kinds of situations.The accuracy rate and recall rate of corner point are both 100%,and the average error is only 1.204×10^(-3) m.The proposed method can achieve accurate cuboid parameterization fitting.

关 键 词：点云 长方体基元 角点提取 参数化拟合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TH744.5[自动化与计算机技术—计算机科学与技术]