i£3BB—Si—jMBäMMBB—MBK
HBasoBBBBBGSKS
r 9K
CMRT09: Object Extraction for 3D City Models, Road Databases and Traffic Monitoring - Concepts, Algorithms, and Evaluation
matches here), but the overall accuracy seems to be better, see
e.g. the height of the buildings. Also the detailed look from ver
tical shows less noise, compared to the two-fold matches before.
5 CONCLUSIONS AND OUTLOOK
This paper reports about the utilization of the dense image mat
ching technique Semi-Global-Matching to a set of high resolu
tion oblique airborne images. The images were acquired from
different platforms and thus in different configurations. The com
parison of the depth maps from matching with a reference com
puted from LIDAR data showed that roughly 70% of all mat
ches are within an error range of ± 3pixel, however, also the
residual errors from camera calibration and orientation have an
impact on this evaluation. The remaining matches can be consid
ered as blunders. How can those blunders be removed and the
noise level be reduced? If multiple overlaps are available, sophis
ticated error analysis prior to triangulation is feasible (Hirsch
müller, 2008). Also the method as applied here shows good re
sults, namely to eliminate wrong matches through linking mat
ches of adjacent images and applying a double check through
a direct match. Other authors use the much stronger trinocular
stereo geometry for matching (Heinrichs et al., 2007), or apply a
similarity criterion for multiple views directly (Besnerais et al.,
2008). If only two-fold overlap is available - as mostly in fa
cade observations from oblique images - one method could be to
incorporate reliable SIFT features within the SGM approach di
rectly: The disparities as defined by them can be used to reduce
the respective matching cost.
The point cloud as resulted from the triangulation of the respec
tive matches revealed the sensitivity to the ray intersection angle
and base length of cameras. For instance in the case of consec
utive FLI-MAP images the theoretic standard deviation of trian
gulated points in viewing (depth) direction is - due to the small
effective baseline - around lm, but perpendicular to that - due
to the low flying height - around 2cm only. In the shown ex
amples these theoretic measures were confirmed. In the tested
images from Pictometry the intersection geometry is better be
cause of the longer baseline. In general, the overall structures on
the building faces are well represented, but the noise reduction
needs further attention.
In the current research the focus is put on the automatic detec
tion and extraction of buildings in oblique images. Here, the
point cloud as derived from the matching can give valuable cues.
Another issue concerns the derivation of a more complete cove
rage by merging the point clouds as derived from different view
ing directions. At least for the roof areas this can be done in
a similar manner as shown above, namely through linking mat
ches, since the majority of roof areas are visible from multiple
directions.
ACKNOWLEDGEMENTS
I want to thank Daniel Oram for providing the code for general
rectification on his homepage 6 . Further I like to thank Matthias
Heinrichs, TU Berlin, for providing me with his code for Semi-
Global-Matching. I also want to thank BLOM Aerofilms for pro
viding the Pictometry dataset. Finally I would like to thank the
anonymous reviewer for their comments.
REFERENCES
Albertz, J. and Kreiling, W., 1980. Photogrammetrisches
Taschenbuch. 3. edn, Herbert Wichmann, Karlsruhe.
Besnerais, G. L., Sanfourche, M. and Champagnat, F, 2008.
Dense height map estimation from oblique aerial image se
quences. Computer Vision and Image Understanding 109(2),
pp. 204 - 225.
Früh, C., Sammon, R. and Zakhor, A., 2004. Automated texture
mapping of 3d city models with oblique aerial imagery. In: 3D
Data Processing Visualization and Transmission, International
Symposium on, pp. 396-403.
Gerke, M., 2008. Dense image matching in airborne video se
quences. In: ISPRS: XXI congress : Silk road for information
from imagery, Vol. XXXVII-B3b, International Archives of Pho-
togrammetry and Remote Sensing, Beijing, pp. 639-644.
Gerke, M. and Nyaruhuma, A., 2009. Incorporating scene con
straints into the triangulation of airborne oblique images. In:
High-Resolution Earth Imaging for Geospatial Information, Vol.
XXXVIII, International Archives of Photogrammetry and Re
mote Sensing, Hannover.
Hartley, R. I. and Zisserman, A., 2004. Multiple View Geometry
in Computer Vision. Second edn, Cambridge University Press.
Heinrichs, M., Hellwich, O. and Rodehorst, V., 2007. Efficient
Semi-Global Matching for trinocular stereo. In: IAPRS, Vol. 36,
pp. 185-190. Part 3-W49A (PIA conference. Munich, Germany).
Hirschmüller, H., 2008. Stereo processing by Semi-Global Mat
ching and Mutual Information. IEEE Transactions on Pattern
Analysis and Machine Intelligence 30(2), pp. 328-341.
Hirschmüller, H., Schölten, F. and Hirzinger, G., 2005. Stereo
vision based reconstruction of huge urban areas from an airborne
pushbroom camera (HRSC). In: Lecture Notes in Computer Sci
ence: Pattern Recognition, Proceedings of the 27th DAGM Sym
posium, Vol. 3663, Springer. Berlin, pp. 58-66.
Höhle, J., 2008. Photogrammetric measurements in oblique
aerial images. Photogrammetrie Fernerkundung Geoinformation
l.pp. 7-14.
Lemmen, M., Lemmen, C. and Wubbe, M., 2007. Pictometry :
potentials for land administration. In: Proceedings of the 6th FIG
regional conference, International Federation of Surveyors (FIG).
Lowe, D. G., 2004. Distinctive image features from scale-
invariant keypoints. International Journal of Computer Vision
60(2), pp. 91-110.
Mishra, P, Ofek, E. and Kimchi, G., 2008. Validation of vec
tor data using oblique images. In: Proceedings of the 16th
ACM SIGSPATIAL International conference on advances in Ge
ographic Information Systems, ACM, Irvine, California.
Oram, D., 2001. Rectification for any epipolar geometry. In:
Proceedings of the 12th British Machine Vision Conference
(BMVC).
Petrie, G. and Walker, A. S., 2007. Airborne digital imaging tech
nology: a new overview. The Photogrammetric Record 22(119),
pp. 203-225.
Seitz, S. M., Curless, B., Diebel, J., Scharstein, D. and Szeliski,
R., 2006. A comparison and evaluation of multi-view stereo re
construction algorithms. CVPR 1, pp. 519-526.
Wang, M., Bai, H. and Hu, F., 2008. Automatic texture acqui
sition for 3d model using oblique aerial images. In: Intelligent
Networks and Intelligent Systems, International Workshop on,
pp. 495^498.
’http://www.aspa29.dsl.pipex.com (accessed 15 March 2009)
82
