改进RANSAC的无人机三维模型建筑轮廓半自动提取  

Semi-automatic contour extraction of building in UAV 3D model based on improved RANSAC

作　　者：车德福[1] 钟壬清 马保东[1] 蒋紫薇 CHE Defu;ZHONG Renqing;MA Baodong;JIANG Ziwei(School of Resources and Civil Engineering,Northeast University,Shenyang 110819,China)

机构地区：[1]东北大学资源与土木工程学院,沈阳110819

出　　处：《桂林理工大学学报》2024年第4期649-655,共7页Journal of Guilin University of Technology

基　　金：国家自然科学基金项目(41871310);中央高校基本科研业务专项(N2124005,N2001020)。

摘　　要：为提取直角类型建筑物的直线轮廓特征,提出了一种无人机倾斜摄影三维模型中建筑轮廓的半自动提取方法。该方法首先通过人工选取建筑物任意墙面上一点,在该点利用射线扫描求交方法获取按照建筑物实际轮廓走向存储的有序点集;再将随机抽样一致RANSAC算法引入点集的直线特征提取中,结合轮廓点的有序性,设置随机抽样的起点与范围,约束共线点的距离阈值,在最优模型求解中引入最小二乘拟合以提取直线;最后对结果中近似直线进行合并,通过相邻直线两两求交确定建筑物轮廓角点及其轮廓线。选取某地倾斜摄影模型中两处建筑物进行轮廓提取验证,结果表明,相对于Hough、直线段检测LSD算法,改进RANSAC算法对简单类型直角建筑物的轮廓提取时间分别加快了18.9%、18.9%,面对复杂类型直角建筑物的轮廓提取时间分别加快了14.5%、24.4%,提取的轮廓在数量、位置、顺序、走向等方面均与实际情况更加吻合,能够有效实现直角类型建筑物的直线轮廓半自动化提取,证明了所提出算法的适用性与可行性。In order to extract the linear contour features of right-angle buildings,a semi-automatic extraction method of building contour from UAV oblique photography of 3D model is proposed in this paper.Firstly,a point on any wall is manually selected,and the ordered contour points of the building arranged according to the actual trend are obtained through the point and ray scanning intersection method.Then,the random sample consensus(RANSAC)algorithm is introduced into the linear feature extraction of point set.Combining with the orderliness of contour points,the starting point and range of random sampling are set,the distance threshold of the collinear points is restricted and least squares fitting is introduced into the optimal model solution to extract counter lines.Finally,the approximate lines in the results are merged by the least square fitting method,and the corners and contour lines of building are determined by the intersection of adjacent lines.In this paper,a building in a certain oblique photography model is selected for verification.The experimental results show that compared with Hough and LSD algorithms,the improved RANSAC algorithm speeds up contour extraction of simple type right angle buildings by 18.9%and 18.9%,and complex type right angle buildings by 14.5%and 24.4%,respectively.And the extracted contour results are more consistent with the actual situation in terms of quantity,location,sequence,direction,etc.,which can effectively achieve the semi-automatic extraction of contours of right-angle buildings and proves the applicability and feasibility of the proposed algorithm.

关 键 词：建筑物轮廓 RANSAC 倾斜摄影 半自动提取 三维模型 

分 类 号：P231[天文地球—摄影测量与遥感]