Fusion of LIDAR Data and Large-scale Vector Maps for
Building Reconstruction
ABSTRACT
LIDAR data contains plenty of height information, while vector maps preserve accurate building
boundaries. From the viewpoint of data fusion, we try to integrate LIDAR data and large-scale
vector maps to perform building modeling. The proposed scheme comprises six major steps: (1)
preprocessing of LIDAR data and vector maps, (2) extraction of point clouds that belong to a
building, (3) construction of the facets from the point clouds, (4) segmentation of planar faces,
(5) determination of 3-D boundaries of buildings, and (6) building reconstruction. In the
preprocessing of stage, the height variation of the aboveground objects is analyzed and the
vertices of a building boundary and the number of stories of the building are obtained. Using the
vertex locations and rough heights, the point clouds that belong to a building can be selected.
Then a triangulated irregular network is built for representing the facets of the point clouds.
Segmentation of planar faces is implemented by considering the angles among surface normal
vectors and the minimum area. After detection for planar roof faces, 3-D roof edges are
determined by intersecting roof planes. Finally, building models are reconstructed by using SMS
method.
1. INTRODUCTION
In response to the development of 3D spatial information for urban planning and management,
cyber city is getting more important. In the cyber city, the information of the real world are
stored and reconstructed digitally in the computer system. The application of the cyber city is
essential, such as in the map revision, change detection, transportation, urban planning,
environmental planning [Danahy, 1999], flight simulations [Volz & Klinec, 1999], etc. Because
the cyber city is the description of the real world, the spatial analysis in the cyber city could be a
reliable reference to city managers.
Building models, among others, could be the most important elements in a cyber city. Thus,
building modeling becomes an important task in photogrammetry and GIS areas. Building
reconstruction could be divided into two parts, i.e., fully automatic approach and semi-automatic
approach. Fully automation is always a substantial goal. Because it is difficult to achieve
automatic interpretation, many researches aim at semi-automatic method [Lang et al., 1995;
Chio and Wang, 1999]. Building model may be accomplished by using aerial photography such
as using CSG model-image fitting [Tseng,2003] and SMS method [Rau & Chen,2002]. Due to
its maturity, LIDAR (light detecting and ranging) has demonstrated profound potentials in
automatic building reconstruction. Many researches for the building reconstruction are based on
LIDAR data in the recent years [Haala & Brenner, 1997; Vögtle & Steienle, 2000; Chen et al.,
2003]. LIDAR data contains plenty of height information, while vector maps preserve accurate
building boundaries. From the viewpoint of data fusion, we try to integrate LIDAR data and
large-scale vector maps to perform building modeling. The workflow of investigation is shown
in figure1.
Fig. 1. Work of investigation
