Automatic planar shape segmentation from indoor point clouds

Abstract

The use of a terrestrial laser scanner (TLS) has become a popular technique for the acquisition of 3D scenes in architecture and design. Surface reconstruction is used to generate a digital model from the acquired point clouds. However, the model often consists of excessive data, limiting real-time user experiences that make use of the model. In this study, we present a coarse to fine planar shape segmentation method for indoor point clouds, which results in the digital model of an indoor scene being represented by a small number of planar patches. First, the Gaussian map and region growing techniques are used to coarsely segment the planar shape from sampled point clouds. Then, the best-fit-plane is calculated by random sample consensus (RANSAC), avoiding the negative impact of outliers. Finally, the refinement of planar shape is produced by projecting point clouds onto the corresponding bestfit-plane. Our method has been demonstrated to be robust towards noise and outliers in the scanned point clouds and overcomes the limitations of over- and under-segmentation. We have tested our system and algorithms on real datasets and experiments show the reliability of the proposed method against existing region-growing methods.