基于球面投影的单站地面激光点云直线段提取方法

摘要

Due to the discrete distribution computing errors and lack of adaptabi l ity are ubiquitous in the current straight l ine extraction for TLS data methods.A 3D straight l ine segment extraction method is proposed based on spherical projection for single station terrestrial laser point clouds.Fi rstly,horizontal and vertical angles of each laser point are calculated by means of spherical coordinates, intensity panoramic image according to the two angles is generated.Secondly,edges which include straight l ine features are detected from intensity panoramic image by using of edge detection algorithm.Thi rdly,great ci rcles are detected from edges of panoramic image using spherical Hough transform.According to the axiom that a straight l ine segment in 3D space is a spherical great ci rcle after spherical projection, detecting great ci rcles from spherical projected data sets is essential ly detecting straight l ine segments from 3D data sets without spherical projection.Final ly,a robust 3D straight l ine fitting method is employed to fitting the straight l ines and calculating parameters of the straight l ine segments.Experiments using different data sets and comparison with other methods show the accuracy and appl icabi l ity of the proposed method.%由于激光点云离散分布于三维空间，导致特征计算困难，目前针对地面激光扫描点云的直线提取方法普遍存在计算量大、提取错误、或适应性不足等问题。本文提出了一种基于球面投影的单站地面激光点云直线段提取方法：首先将三维点云投影到球面上，球面投影后的三维点云保持了正确的邻接关系，同时也解决了三维数据计算量大的问题；然后，利用球面霍夫变换，正确提取球面投影后点云中的直线段特征；最后，利用稳健的直线拟合算法计算点云中的三维直线段参数。通过对室内室外场景数据的直线段提取试验及与商业软件 Geomagic Studio 等方法的对比试验，验证了本方法的正确性、适用性和准确性。