inbul 2004
). A more
1 be found
1? problem,
Model was
; Zo, D, Q,
j)bject and
g problem
| (b) that
ailable. In
> matching
| about the
| approach
ollinearity
3D object
h (as used
ons cannot
emma and
parameters
is usable.
1e exterior
St squares
' shown in
(4)
(5)
L,-L;, can
offers the
rs (Xo, Yo
33), which
ed spatial
od is the
the spatial
umber of
even DLT
es for the
erive the
be set up
ast square
does not
matrix. A
ze, 1988)
te of an
Jarantee a
eters only
errors are
npute the
on.
different
ire 7. The
a suitable
ways are
ess can be
ntation is
here is no
; a semi-
extracted
ually co-
| the DLT
e exterior
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
orientation, which act as input data for the feature based spatial
resection. The other way is to start the process if initial values
are existing. In that case we have good basic information, hence
we can extract the necessary features more specific. To identify
the object and regions of interest for feature extraction a
“General Hough Transform” is utilized. Once the regions are
known and the features extracted, the spatial resection algorithm
is started to compute the exterior orientation parameters.
Input: collected image ]
Input: City model ]
Input: rough exterior orientation
General Hough Transformation
(extraction of areas of interest)
y
manual feature extraction
(whole image)
automatic feature extraction
(whole image)
automatic feature extraction
(in areas of interest)
Searching for
corresponding
semi automatic
matching
semi automatic
matching
image features
and model
features
DLT DLT
(feature based) (feature based)
Y
( Spatial Resection (feature based) |
Exterior Orientation
Figure 7. Process for determination of exterior orientation
3. EXPERIMENTS AND RESULTS
Several examples were choose to investigate the feasibility of
our approach. The investigations are based on a 3D CAD
dataset of the city of Stuttgart provided by the City Surveying
Office. This 3D CAD city model was created by manual
photogrammetric stereo measurements (Wolf, 1999). In the
dataset of the City Surveying Office a large amount of detail is
available and its accuracy is high. Therefore we decided to use
this dataset. Additionally also a synthetic dataset was prepared
to be able to study the quality of the feature based method. In
the synthetic dataset an ideal camera was simulated for
projection of objects into image space. The advantage of the
simulated dataset is that there are no distortions in the image
Space, which offers the possibility to study the quality of the
feature based spatial resection method.
real dataset:
rough exterior orientation (collected data)
X |910.62 m € | 10.63?
Y |89.17m o | 64.7?
Z 158.15m K | 0.0° (predefined)
spatial resection
points straight lines differences
X | 904.45 m 904.66 m AX] | 0.21 m
Yl7213m 72.4] m |AY] 0.28 m
Z | 52.46 m 52.28 m IAZ| 0.18 m
907
o [ 9.9 10.1? le | 0.27
q | 61.4? 63.0? Ap] | 1.6?
Kk | -2.56? -3.4? [Ax 0.84°
Table 1. Results of the real dataset
Table 1 shows results using a real dataset and Table 2 shows
result of a synthetic dataset. In the table of the real dataset the
rough exterior orientation collected by the prototype sensor is
displayed. These values are also used as initial values in the
spatial resection process. The results displayed in both tables
are created by the feature based spatial resection method as well
as by a spatial resection using manually selected tie points as
input data. The results of the point based method act as
reference data, as this method provides the most accurate
results. In the table of the synthetic dataset (Table 2)
additionally the results of the DLT method are added, to show
their ability for determining initial values.
synthetic dataset:
rough exterior orientation (predefined/synthetic)
X {905.00 m o | 10.0?
Y |72.00m © | 64.0°
Z | 51.00m K y 40?
DLT - straight lines (for initialisation)
X | 903.49 m eo 9383?
Y {7123 nm © | 64.61°
Z \51 lim x |-0.03?
spatial resection
points straight lines differences
X | 905.039 m 905.002 m [AX] | 0.037 m
Y | 72.012 m 72.032 m JAY] | 0.02 m
Z | 51.0871 m 51.0796 m AZ} | 0.008 m
c | 9.908? 9.9199 Ao | 0.011?
o | 63.978? 64.020° jAg| | 0.042?
K | -0.018° -0.009° jAx| | 0.009°
Table 2. Results of the synthetic dataset
Comparing the results in Table 1 (real dataset) we can see that
the "straight lines" method compared the point based method
provides nearly the same result. Position differences are in the
order of — 20 cm and differences for the orientation angles are
in the order of 1-2 degree. As the point based estimation
represents the optimal result, we can conclude that the errors
occur by inaccuracies in the extraction of straight lines affected
by the image quality and image resolution, or by inaccuracies in
the distortion parameters. Considering the synthetic dataset in
Table 2 which uses the same image coordinates for the point
based method as well as for the straight line based method, we
can see nearly the same result for the exterior orientation. This
shows that the straight line based method is suitable to
determine reliable results for the exterior orientation
parameters.
4. CONCLUSION
In the article we have described the idea of the NEXUS platform,
which offers the possibility to represent the real world as a
world model. We have pointed out the key problem of LBS:
“Fist the system has to determine where the mobile user is
located at, than the system can provide support to the user.”
Based on this key problem we have described general methods
for location sensing and also a method, which uses scene
analysis. For the method based on scene analysis we have
presented an implementation of a process for estimating
