This paper presents a method that aims at reconstructing a 3D building from point clouds measured by 3D scanner. It starts from the idea that it is easier to rebuild a scene using available knowledge about the scene\’s elements. This solution has to consider the three following aspects. How to find objects in a cloud of points? How to define a geometric and semantic coarse model? Which algorithms to use as a propagation method to find all objects in the cloud of points? In our solution the user has to assign the context by defining a coarse model of the building to be reconstructed. Then the user interactively selects a set of points in the cloud that represents an element. The selection is also mapped to the coarse model by assigning the corresponding wall in the \“CM\”. Then the user starts the reconstruction algorithm. Within an iterative process the plane representing the wall is found and will be used to correct the model. The process starts with the mapped plane, corrects it, and continues with information in \“CM\” to detect an adjacent plane by propagation.

} }