Towards automatic indoor reconstruction of cluttered building rooms from point clouds
Terrestrial laser scanning is increasingly used in architecture and building engineering for as-built modelling of large and medium size civil structures. However, raw point clouds derived from laser scanning survey are generally not directly ready for generation of such models. A manual modelling phase has to be undertaken to edit and complete 3D models, which may cover indoor or outdoor environments. This paper presents an automated procedure to turn raw point clouds into semantically-enriched models of building interiors. The developed method mainly copes with a geometric complexity typical of indoor scenes with prevalence of planar surfaces, such as walls, floors and ceilings. A characteristic aspect of indoor modelling is the large amount of clutter and occlusion that may characterize any point clouds. For this reason the developed reconstruction pipeline was designed to recover and complete missing parts in a plausible way. The accuracy of the presented method was evaluated against traditional manual modelling and showed comparable results.