Geoscience Reference
In-Depth Information
automatic methods of processing are used with increasing amounts of data that must
be processed in shorter time periods. Current automatic techniques are still under
development and can only partially exploit the potential of the data. The real
potential of the data is usually unused. This issue is therefore a challenge for experts
from a wide range of disciplines like remote sensing, photogrammetry or computer
vision.
The most common form of spatial elevation data is a point cloud which can be
obtained by active sensors like airborne laser scanning (ALS) systems which use
the Light Detection and Ranging (LiDAR) principle. The spatial data can be also
derived by image matching techniques using satellite or aerial images. Point cloud
data represent the surface geometry by an object independent distribution of points
with uniform quality, however, this form of representation is not appropriate for
many applications. For more sophisticated tasks, a generalization and simpli
cation
of the digital surface model (DSM) is necessary. The generation of 3D building
models is just such the case. The
fields of application of 3D building models are
quite various such as visualizations, urban planning, environmental monitoring (for
example air pollution, propagation of road traf
c noise etc.), propagation of elec-
tromagnetic waves for telecommunication applications and the generation of
fl
ood
maps.
Image matching techniques are very popular and progressive methods nowa-
days, however, they are extremely dependent on the quality of input aerial images.
Ground sampling distances (GSD) and overlaps between images have a major
impact on the acquisition of high quality outputs. Most of the presented results were
typically achieved by high quality input datasets with large forward and side
overlaps between images and GSD under 10 cm. Data with these speci
cations can
be collected relatively easily from small areas, but it seems that it will not be
economic to collect such data for large areas for entire countries. This paper gives
a review of the possibility of fully automatic generation of 3D building models from
point clouds covering large areas. For this reason, point clouds collected only by
airborne laser scanning will be used as input data in this paper.
2 Related Work
The
cation of points
describing the buildings. These points represent mainly building roofs and their
parts in case they were collected from the air. The second important task is the
segmentation and conversion of these roof parts into geometrically and topologi-
cally correct building models. These tasks are really challenging and most scienti
first important task in 3D building reconstruction is the identi
c
papers solve this problem using other external sources of information like maps or
even better ground plans. Building ground plans can be obtained from digital
cadastral maps or GIS layers containing building footprints. This information can
rapidly help for building reconstructions as no sophisticated algorithms must be
used for the classi
cation of raw point cloud [ 1 ]. The buildings are reconstructed on
Search WWH ::




Custom Search