This paper describes a consensus‐set estimation for building roof‐plane detection using a stratified random sample consensus (sRANSAC) algorithm applied to point clouds acquired by laser scanning systems. The main idea is to use one initial classification to generate consensus‐set candidates to optimise the sampling mechanism compared to the original RANSAC. The initial classification is performed using mathematical morphology to filter ground returns and estimate local variance information to detect potential planar regions. Thus, the algorithm can prioritise points within planar segments and the number of iterations can be estimated dynamically from available data. The results based on experiments using five different lidar datasets indicate that the proposed method reduces the number of computations for building roof‐plane detection and also improves accuracy compared to RANSAC.

中文翻译：

---

分层随机样本共识提取建筑物屋顶平面

本文介绍了使用分层随机样本共识（sRANSAC）算法对建筑物屋顶平面进行检测的共识集估计，该算法应用于激光扫描系统获取的点云。主要思想是使用一种初始分类来生成共识集候选者，以与原始RANSAC相比优化抽样机制。使用数学形态学执行初始分类，以过滤地面收益并估算局部方差信息以检测潜在的平面区域。因此，该算法可以对平面段内的点进行优先排序，并且可以从可用数据中动态估算迭代次数。