AUTOMATIC 3D MODELLING OF INDOOR MANHATTAN-WORLD SCENES FROM LASER DATA

We developed an automatic technique for the reconstruction of Manhattan-world interior scenes from point clouds. Our method mainly focuses on the reduction of human intervention in the modelling process, thus it aims to a full automation of the modelling phase with a consequent optimization of the efficiency of the laser-scanning project pipeline. We refer to a reconstruction complexity typical for indoor scenes. The work flow starts with a volume sweep reconstruction of the interior from the three-dimensional point cloud. As result of a discrete translational plane sweep, the input data is segmented into separate point sets including the floor, ceiling and wall points. Consequently, each point is assigned to a surface of the volume. The ground plan contours are extracted with a cell decomposition approach after partitioning the floor surface into rectangular cells of variable size. Only cells considered suitable are added to the ground shape and unified to define the ground plan. Along the ground plan contour, the walls are raised from the floor up to the ceiling level. Finally, the interior model is enhanced by the addition of built-in feature like doors.