基于边缘特征的港口场景点云配准方法  

Point cloud registration method of port scenarios based on edge features

作　　者：徐承军[1] 吴志意 Xu Chengjun;Wu Zhiyi

机构地区：[1]武汉理工大学交通与物流工程学院,武汉430063

出　　处：《起重运输机械》2024年第20期67-75,共9页Hoisting and Conveying Machinery

摘　　要：传统ICP等配准算法对初值敏感容易陷入局部最优,难以在港口等较大场景点云配准的过程中应用。文中针对这一问题,借助港口场景及设备包含丰富的线面几何特性的特点,提出一种基于点云边缘直线特征的点云配准方法。首先,通过区域生长和区域合并将点云分割成平面,根据平面投影得到二维图像后利用二维图像轮廓提取和最小二乘法拟合得到二维线段,然后将2d线段重投影至三维得到3d线段点。将提取到的边缘直线采用先粗后精的配准方法,先使用Super4PCS粗略对齐2帧点云边缘线,最后使用改进Trimmed-ICP对初始点云实现精确配准。通过实地采取港口场景点云数据,通过实验证明文中方法的实用性和优越性。Registration algorithms,including traditional ICP,are sensitive to initial values and easily lead to local optimum,which makes it difficult to be applied them in large-scale point cloud registration such as ports.Considering the above situation,a point cloud registration method based on the line feature of point cloud edge is proposed according to the fact that port and equipment contain rich geometric characteristics of line and surface.Firstly,the point clouds were divided into planes by region growing and region merging,and two-dimensional images were obtained by plane projection,and then two-dimensional line segments were obtained by contour extraction and least square fitting.Then,two-dimensional line segments were reprojected to three-dimensions to get three-dimensional line segment points.The extracted edge lines were processed by coarse registration before fine registration.Firstly,the edge lines of two point clouds were roughly aligned by Super4PCS.Finally,the initial point clouds were accurately registered by improved Trimmed-ICP.Experiments with collected point cloud data of port on the spot proved the practicability and superiority of the proposed method.

关 键 词：大场景点云 线特征 点云分割 点云配准 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]