Preparation and enhancement of 3D laser scanner data for realistic coloured BIM models

摘要

Realistic point cloud models are frequently required in order to create efficient 3D data processing algorithms. In a building information modelling context, for example, the segmentation and object recognition of point clouds become extremely difficult tasks when the inputs of the algorithms are not sufficiently good. This paper proposes a method with which to efficiently solve two of the most important issues during the creation of realistic 3D data with laser scanners: the treatment of specularity in coloured point clouds and the view merging process. We particularly deal with wide-range laser scanners with the objective of generating realistic orthoimages of the main structural elements of the insides of buildings (i.e. walls, ceilings and floors). The new algorithm gathers the coloured points belonging to structural elements and delimits the specular regions originating from non-controlled or directional lighting sources (e.g. flashes). The restoration of these specular regions is solved in a subsequent stage in which several partial views of a structural element are integrated into a unique orthoimage. Finally, the quality of the resulting orthoimage is evaluated by comparing it with the corresponding ground truth image. This method has been tested on a public database, yielding encouraging results.