measurements are sufficiently acquired over buildings (see
figure 13 (a)). However, as lidar points are acquired with
less point density over a building, more errors are produced
around its boundaries (figure 13 (b) and (c)). This is because
the detection of data-driven lines and model-driven lines is
more difficult over a building with coarser point density than
the one with denser point density. As a result, mis-location
of data-driven lines and model-driven lines leads to the
generation of delineation errors around building boundaries.
E Et » a
Figure 13. Building delineation errors; the first column
shows cut-out Ikonos images overlaid with
building-labelled lidar points; the second column
shows building delineation errors
(c) UCL building map — (d) OS MasterMap
Figure 14. Reference data errors
Reference data error: these errors are caused by the
inherent faults in the OS MasterMap” (see yellow coloured
pixel in figure 12). As can be seen in figure 14, the UCL
building map can successfully delineate boundaries of a
building based on the result of lidar measurements and
Ikonos image. However, the OS building map missed some
part of that building (cf. figure 14 (c) and (d)). As outlined
earlier, this error caused by a time difference between the
acquisition of the Ikonos image and lidar data, and the
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
construction of the OS data. The analysis of the reference
errors suggests that the developed building extraction
technique can be also used for applications detecting
changes in an urban environment and supporting map
compilation.
7. CONCLUSIONS
This paper presented a system for automatically detecting
building objects and delineating their boundaries from
Ikonos images and lidar data. A few new ideas to combine
complementary nature of intensity images and high-quality
of 3D information to solve problems associated with
building detection and building description are introduced.
The overall success of the developed building extraction
system was evaluated in comparison with the OS
MasterMap" ground plan. The results highlights Ikonos
images can be used in topographic mapping at large scale in
a combination of lidar data. The current system is limited to
delineating polygonal shapes of buildings with flat roofs.
Thus, a further development must be directed to reconstruct
3D roof structures based on the ground plans extracted by
the current techniques.
REFERENCES
Burns, J.B., Hanson, A.R. and Riseman, E.M., 1986.
Extracting straight lines. [EEE Pattern Analysis and
Machine Intelligence, 8(4):425-455.
Fuchs, H., Kedem, Z.M. and Naylor, B.F., 1980. On visible
surface generation by a priori tree structures. Computer
Graphics, 14(3):124-133.
Holland, D., Guilford, B. and Murray, K., (eds) 2002.
OEEPE-Project on topographic mapping from high
resolution space sensors. Media-Print, Paderborn.
Holland, D. and Marshall, P., 2003. Using high-resolution
satellite imagery in a well-mapped county. Proceedings of
Joint ISPRS Workshop on High Resolution Mapping from
Space 2003, Hannover, Germany, 6-8 October 2003. CD-
ROM.
Kim, J.R. and Muller, J-.P., 2002. 3D reconstruction from
very high resolution satellite stereo and its application to
object identification. Proceedings of ISPRS technical
Commission 1V Symposium 2002 on Geospatial theory,
Processing and Applications, Ottawa, Canada, July 9-12,
2002, 34(4). CD-ROM.
Lee, D.S., Shan, J. and Bethel, J.S., 2003. Class-guided
building extraction from lkonos imagery. Photogrammetric
Engineering & Remote Sensing, 69(2):143-1 50.
Shufelt, J.A., 1999. Performance evaluation and analysis of
monocular building extraction from aerial imagery. IEEE
Pattern Analysis and Machine Intelligence, 21(4):31 1-326.
Sohn, G. and Dowman, 1.J., 2002. Terrain surface
reconstruction by the use of tetrahedron model with the
MDL criterion. In: Proceedings the Photogrammetric
Computer Vision, Graz, Austria, September, pp.336-344.
KEY
ABST
At pre
acquis
usuall
edge,
The a
urban
recons
compx
plane
estime
Exper
rotatic
a prec
The c:
task: |
correc
the Gl
as we
is per!
in flig
over |
acquis
and th
usuall
the co
data.
Differ
atic er
2001).
pear:
source
a loca
depen
the ot
vertic:
The g
help fi
forma
registi
The m
surfac
A pre
with a
a clas:
new Ir