The automatic reconstruction of three-dimensional models of buildings has been for several years a major research project. Models of buildings are increasingly used for economic reasons in spatial planning. Among the global methods of detection, useful is the active measurement system ALS. However, building reconstruction based only on a cloud of points is complicated, therefore it is very useful to use additional information from other sources. In the present study were used cadastral maps, which allowed a whole set of points to choose only those which reflected from the roofs of buildings. Vectors from the ground plans were obtained as the edges of the roofs of buildings, which were also used as boundaries based on cut points that represented the roof. Roof surfaces were detected from the LIDAR data. The shape of a roof is automatically determined using the algorithm based on the split merge method. The aim of the algorithm is the transition from particular to general, so it is possible to detect the size of roof elements, which correspond to the size defined at the beginning voxels. The process of identifying roofs runs iteratively, based on the parameters describing the plane. Vectors are raised to the height of the building roof. Models of buildings were derived through integration of vector data from the ground plans, and data from airborne laser scanning is placed on digital terrain models. A terrain model was built of automatically filtered clouds of points.