Refining Existing 3d Building Models with Terrestrial Laser Points Acquired from a Mobile Mapping Vehicle

This paper presents a method to automatically refine existing 3D building models with laser points acquired from a terrestrial mobile mapping system (MMS). 3D models derived from aerial imagery are in general less accurate than 3D information derived from terrestrial laser scanning. Therefore, point clouds are used to refine these models in order to make them precisely correspond to the scanned façade geometry. After a preliminary global registration using the ICP algorithm, each visible façade is individually processed and compared to a filtered point cloud containing only the façade points. First, an individual registration fits the façade plane to the laser points. Then the façade corners are detected in the point cloud by analysing changes in the gutter height. As façade contours can be skewed by lateral cornices, the latter are also detected to refine delineations. Finally, façade corners are adjusted in order to restore original connections between buildings. As a result, each façade of the model is accurately positioned and delineated thanks to the laser point cloud, whilst respecting the structure and the topology of the original model. The façade number and verticality are preserved. This algorithm was tested on several data sets from different MMS and showed satisfactory results, without any preliminary adjustments from the user.