基于多结构估计的建筑物激光雷达点云特征线提取算法

Extract Feature Lines from Building LiDAR Point Cloud Based on Multi-structure Estimators

建筑物激光雷达(light detection and ranging,LiDAR)点云特征线对于多视角点云配准、建筑物对称性检测、建筑物三维重建等应用具有十分重要的意义.由于LiDAR点云具有数据量庞大的特点,传统的算法难以实现建筑物特征线的快速提取.针对这个问题,提出一种基于多结构鲁棒估计的建筑物特征线提取算法,该算法利用历史模型信息进行条件采样,并通过迭代搜索符合所有特征线性质的模型.根据建筑物LiDAR数据的实验结果表明,该方法与传统的RANSAC(random sample consensus)、MLESAC(maximum likelihood estimation sample consensus)等算法相比,避免了无效、重复的特征线采样过程,在相同时间内可获取更多的直线内点,从而有效提高了建筑物特征线的提取效率.

Feature lines extracted from building LiDAR（Light Detection and Ranging）point cloud data are of great significance in multiple views registration,building symmetry detection,3Dsurface reconstruction,among others.Since the LiDAR data are generally associated with a huge amounts of 3Dpoints,traditional algorithms suffer from the time complexity of rapidly extracting feature lines from building point cloud.In order to solve this problem,we present a feature lines extracted algorithm based on multi-structure robust estimation.In the proposed method,historical models generated by random strategy have been used for conditional sampling new models.Consequently,the searching process aims at extracting all feature lines from the model set.In the section of experiments,the multi-structure algorithm has been compared with the RANSAC（random sample consensus）and MLESAC（maximum likelihood estimation sample consensus）.Results acquired from our LiDAR dataset indicate that the proposed method improves the efficiency of building feature lines extraction,since the multi-structure algorithm avoids many invalid and repeated sampling processes.Therefore,we can generate more feature lines at the same time.