In: Stilla U, Rottensteiner F, Paparoditis N (Eds) CMRT09. IAPRS, Vol. XXXVIII, Part 3/W4 — Paris, France, 3-4 September, 2009
Figure 12. Combination of DTM and building layer to a final
DSM, covering the high resolution study field.
6. GEOMETRIC ACCURACY ANALYSIS
A dataset of 35 check points, measured with GPS and evenly
distributed over the study area, is used to check the geometric
accuracy and quality of the extracted models from stereopair
and triplet. It concerns independent ground control points,
meaning that they are not used in the photogrammetric
processing of the models. Check points are preferred because of
the lack of a more accurate reference surface model. Besides,
the uncertainty of height errors in a reference map is much
bigger than for discrete measured values in the field.
Comparison of a measured height value and the calculated
height value in the model at a certain location gives statistical
information about the accuracy by which reality is modelled.
Some calculated statistics, quantifying the geometric accuracy
are presented in table 2. Distinction is made between the
standard stereoscopic and tri-stereoscopic approach. The a
priori geometric accuracy reflects the quality and robustness of
the image orientation. RMS error in X, Y and Z is given for the
total of 17 ground control points that were used to fix the
mathematical relationship between image and object coordinate
space. For X and Y, sub-pixel accuracy is obtained in both
approaches. RMSE for the Z component is less than 3 pixels. 35
independent check points are used to calculate the RMS error
for Z and the mean Z difference between measured and
calculated value by the model. For both statistics the value is
less than 3 pixels.
A priori geometric accuracy
DSM geometric accuracy
Imagery
No. of GCP RMSX (m) RMSY (m) RMSZ (m)
No. Of CP RMSZ (m) Mean dZ (m)
Stereoscopic
tri-stereoscopic
17 0.68 0.72 2.44
17 0.79 0.78 2.36
35 2.61 2.21
35 2.47 2.06
Table 2. Geometric accuracy analysis.
Visual analysis of the models shows big improvements of the
quality for the surface model derived from the Ikonos triplet.
Noise is reduced and smoothing effects of man-made object are
reduced to a minimum, however the improvements do not
reflect in the quantitative accuracy check. The RMSE and mean
values are slightly better for the triplet than for the stereopair.
This is due to the fact that the improvements are mainly situated
around buildings and other steep changes in elevation. Check
points are mostly measured in open terrain so that they are
clearly identifiable on the imagery. Within these non-complex
areas the surface model from the stereopair gives also optimal
results. To have a better quantification of the improvements,
future work should involve the collection of rooftop heights for
a set of buildings and comparison between the collected ground
truth and the produced models.
7. CONCLUSION
In this treatise an approach is proposed to extract an urban
surface model in a semi-automatic way directly from multi-
scopic Ikonos imagery, in contrast to surface models derived
from manual plotting of building rooftops. The input of the
operator during photogrammetric processing is reduced to a
minimum. Interesting advantages are that it is less labor-
intensive and that the outcome is independent from human
interpretation. Off course manual plotting of buildings will lead
to a higher accuracy and more detailed information, but this task
is very time consuming and will not be cost-effective in some
situations. As from the perspective of the geometric accuracy, as
from the visual analysis we can conclude that the outcome is
encouraging and that acceptable results are reached. At different
levels of the photogrammetric processing of the imagery, efforts
are done to cope with the complexity of modeling an urban
environment. Occlusion and consequently mismatches are
reduced by combining the redundant information of a third
image with a stereopair. Radiometric and geometric
dissimilarities between the multi-temporal imagery are
diminished by preprocessing the individual images.
Combination of three different matching algorithms gives
redundancy and geometric constraints leading to dense and
reliable matching results. Finally, spatial filtering is applied on
the height values of the DSM to reduce smoothing effects and
enhance global DSM quality.
REFERENCES
Baltsavias E., 1991. Multiphoto geometrically constrained
matching. PhD Dissertation, Report No.49, Institute of Geodesy
and Photogrammetry, ETH Zurich, Switzerland.
Baltsavias E., Pateraki M., Zhang L., 2001. Radiometric and
geometric evaluation of Ikonos GEO images and their use for
3D building modelling. Proc. Joint ISPRS Workshop High
Resolution Mapping from Space 2001, Hannover, 19-21
September, 2001.
Baltsavias E., Zhang L., Eisenbeiss H., 2006. DSM generation
and interior orientation determination of ikonos images using a
testfield in Switzerland. Photogrammetrie, Fernerkundung,
Geoinformation, (1), pp. 41-54.
Bethel J.S., McGlone J.Ch., Mikhail E.M., 2001. Introduction
to Modern Photogrammetry, John Wiley & Sons, Inc., New
York, 477 p.
