Automated methods for indoor point cloud preprocessing: Coordinate frame reorientation and building exterior removal

Raw indoor datasets acquired from laser scanning may be tilted and contain outliers, which negatively influence the accuracy of subsequent tasks. Therefore, an automatic framework with three steps is provided to preprocess indoor point cloud. First, a hybrid segmentation method is proposed to extract planar primitives from noisy points. Then, a nonlinear optimization is designed to discover the dominant orientations of the building with topological constraints. Finally, the outdoor points are removed through a novel graph-cut formulation. The experiments on six various buildings indicates that our methods successfully reorient the point cloud and preserve the indoor points. In comparison with other methods, we achieve superior performance on visual inspection and evaluation metrics. Besides, our work only relies on the geometric position and is not limited by Manhattan-World assumption. We present a reliable solution for indoor point cloud preprocessing that have potential for indoor modelling, floorplan generation, and objects detection. •A robust framework is developed for automatic preprocessing indoor point cloud under complex environment.•A hybrid segmentation approach is proposed to enhance the extraction of planes from cluttered point cloud.•A nonlinear optimization model is created to make buildings well-aligned with the coordinate axes.•An energy minimization formulation is designed to remove points outside the building.