Efficient plane extraction using normal estimation and RANSAC from 3D point cloud

•An effective method for extracting indoor scene plane is proposed.•The weighted PCA method is used to estimate the point cloud normal vector more accurately and to avoid noise interference.•Proposed an improved RANSAC method, which can extract the desired plane better and more effectively in complex environments (such as multiple planes crossing, overlapping, etc.). Indoor plane extraction on point cloud has always been a research hotspot, in which random sample consensus (RANSAC) is known as a common algorithm. However, impacted by numerous occluded objects in the interior scene, the point cloud generated by the sensors may be missed in part of the aircraft area. Moreover, the conventional RANSAC method will cause the plane being incorrectly extracted. In this study, an indoor plane detection method is proposed based on space decomposition and an optimized RANSAC algorithm. In this method, the weighted PCA method is exploited to estimate the normal vector from point cloud, then the angular clustering is employed to divide the interior space for obtaining the building components. Subsequently, an optimized RANSAC method is adopted to detect planes from the building components obtained. To be specific, the proposed RANSAC method selects the candidate points by using a heuristic search strategy, and then the mentioned candidate points are used to estimate the final plane. The proposed method can handle the overlapping patches that cannot be extracted by using the conventional RANSAC method. The proposed method is assessed on 4 indoor datasets. As indicated by the experimental results, the proposed method can detect the plane structure efficiently and effectively.