The acquisition of 3D city and landscape models has been a topic of major interest within the photogrammetric community and a number of algorithms became available both for the automatic and semiautomatic data collection. These 3D models are mainly used for visualisation to generate realistic scenes and animations e.g. for planning and navigation purposes. Since area covering urban models are usually extracted from airborne data, a refinement is required especially if realistic visualisations from pedestrian viewpoints are aspired. Of particular interest are algorithms that allow for a fully automatic facade reconstruction process at different levels of detail. For this purpose, in our approach terrestrial LIDAR data as well as facade imagery is used to increase the quality and amount of detail for the respective 3D building models. The a priori information which is provided by the models from airborne data collection is integrated both for the automatic georeferencing of the terrestrial data and the subsequent geometric refinement. After alignment of the terrestrial data, window structures are determined approximately from the LIDAR point clouds and then further refined by 3D edges which are extracted from the overlapping facade images. Our modelling process applies a 3D object representation by cell decomposition, which can be used efficiently for building reconstruction at different scales.