面向三维扫描仪点云与全景图像映射关系的快速建立方法  

Fast establishing method for mapping relationship between 3D scanner point cloud and panoramic image

作　　者：张旭 毛庆洲 时春霖 施以旋 ZHANG Xu;MAO Qingzhou;SHI Chunlin;SHI Yixuan(School of Remote Sensing and Information Engineering,Wuhan University,Wuhan 430079,China;Hubei Luojia Laboratory,Wuhan 430079,China;Troops 61206,Beijing 100042,China)

机构地区：[1]武汉大学遥感信息工程学院,湖北武汉430079 [2]湖北珞珈实验室,湖北武汉430079 [3]中国人民解放军61206部队,北京100042

出　　处：《浙江大学学报（工学版）》2024年第7期1315-1325,共11页Journal of Zhejiang University：Engineering Science

基　　金：国家重点研发计划资助项目(2023YFC3009400,2023YFB2603702)。

摘　　要：针对地面三维扫描仪获取彩色点云时传感器外参数标定过程复杂的问题,提出直接建立点云与全景图像映射关系的方法.提出基于一维最大熵的改进Zernike矩亚像素边缘提取算法,自全景图像中定位靶球,根据三维几何特点从点云中提取靶球.将提取结果作为配准基元,在空间球坐标中构建基元三角形,通过最小角距差法完成基元配对,建立点云与全景图像的初始映射关系.针对图像局部畸变导致的映射偏差,提出基于改进Levenberg-Marquardt算法和自由形变法组合的混合算法逐像素优化数据间的映射关系.利用多种场景的实验数据验证所提方法的可行性.结果表明,标靶自点云和图像中的提取率高,被点云和图像同时识别的标靶利用最小角距差法均能够成功配对.改进Zernike矩相较于传统Zernike矩提取的标靶初始映射误差降低了61.1%;经混合算法优化后,点云与全景图像的映射误差约为1 pixel,数据映射结果稳定且不受测站位置和点云密度的影响.A method was proposed to directly establish the mapping relationship between point clouds and panoramic images in response to the complex calibration process of sensor external parameters when obtaining color point clouds using the ground 3D scanner.Firstly,an improved Zernike moment sub-pixel edge extraction algorithm based on one-dimensional maximum entropy was proposed to locate the target sphere from the panoramic image,and the target sphere was extracted from the point cloud according to the 3D geometric characteristics.Then,the extraction result was used as the registration primitive,and the primitive triangle was constructed in the spatial spherical coordinate.The primitive pairing was completed by the minimum angular distance difference method,and the initial mapping relationship between the point cloud and the panoramic image was established.Finally,aiming at the mapping deviation caused by local image distortion,a hybrid algorithm based on improved Levenberg-Marquardt and free-form deformation combination was proposed to optimize the mapping relationship between data pixel by pixel.The feasibility of the proposed method is verified by the experimental data of multiple scenarios.The results show that the extraction rate of the target from the point cloud and the image is high,and the target recognized by the point cloud and the image is successfully paired by the minimum angle difference method.Compared with the traditional Zernike moment extraction target, the initial mapping error was reduced by 61.1% for the improved Zernike moment. After the optimization of the hybrid algorithm, the mapping error between the point cloud and the panoramic image was about 1 pixel, and the data mapping result was stable and not affected by the position of the station and the point cloud density.

关 键 词：三维扫描仪 全景图像 一维最大熵 映射关系 配准误差 

分 类 号：P232[天文地球—摄影测量与遥感]