CIPA 2003 XIX th International Symposium, 30 September - 04 October, 2003, Antalya, Turkey 
200 
MDS 
(pair) 
Film 
Agfapan APX 
100 
Kodak Technical Pan 
all points 
status = I 
Ml-2 
0.055 
0.044 
0.031 
M2-3 
0.060 
0.043 
0.036 
M3-4 
0.031 
0.055 
0.033 
M4-5 
0.042 
0.031 
0.033 
M6-7 
0.025 
0.016 
0.016 
M7-8 
0.025 
0.017 
0.015 
M8-9 
0.030 
0.010 
0.009 
M9-10 
0.022 
0.028 
0.019 
M10-11 
0.034 
0.023 
0.020 
Mean 
0.036 
0.030 
0.023 
Table 1. RMSE (m) values for the individual DSM and the 
average RMSE from the group for each type of film. The 
right column shows the improvement after the elimination 
of points with r<0.85 (See 4.3.5). 
Figure 8 shows a graphic representation of the DSM 
corresponding to the Ml-2 (APX 100) pair. The grey tones 
respond to the depth value (z). 
Figure 8. Example of DSM constructed with Ml-2 (Pan) pair. 
There is an empty area (hidden) to the right of the tower. 
4.3.5 Correlation coefficient value. For the detection of tie 
points, it is necessary to define a threshold value that works as a 
filter. In the construction of the DSM the possibility of selecting 
a threshold value does not exist. The result is formed by a group 
of points calculated with variable correlation values. The 
application assigns a value (“status”) to each point according to 
the correlation coefficient r: 1: r > 0.85; 2: 0.85 > r > 0.70, 3: 
0.70 > r > 0.50. Status 4 (isolated) and 5 (suspicious) are not 
defined in relation to r. 
The tests of error control show that the elimination of points 
with a worse correlation improves the results. Conserving the 
points with a status of one exclusively improvements are 
obtained in almost every case, especially in those that had 
demonstrated worse precision (right column in Table 1). This 
test shows that the correlation coefficient may be used as an 
estimator of the data reliability. 
4.4 Improvement: a synthesis DSM 
It has already been indicated that the DSMs are constructed 
from a sole stereoscopic pair. Since the application does not 
construct a DSM by using all the data on the network, a 
synthesis has been carried out which follows the steps hereafter: 
• Elimination of gross errors and hidden points in each 
DSM 
• Elimination of all points with status * 1 (not optimum 
correlation) 
• Union of individual DSMs 
• Error control in the synthesis DSM 
The results are clear: the synthesis DSM shows a better 
precision than the individual DSM. In this case, the film is a 
differential factor: the DSM constructed with Agfapan APX 
100 is made up of 565926 points (RMSE = 0.020 m) while with 
Kodak Technical Pan 1057338 points are arrived at (RMSE = 
0.015 m). That is, not only the error value improves at 25 %, 
but also the quantity of points is almost double with Technical 
Pan film. The improvement with respect to the group of DSM 
(Table 1) is important in both cases since the RMSE is reduced 
to half approximately. 
4.5 Discussion 
Convergent networks with multiple images still cannot be 
treated with maximum results using current commercial 
software. In the case of Orthobase Pro, it is possible to obtain 
adjustment statistics for a converging network but it is only 
possible to obtain DSM per image pair. The adjustments worsen 
with the convergence angle. This fact suggests that the 
converging network designs have not been used but that 
adaptations have been made with algorithms of normal 
geometry. 
There is another evidence that indicates that the processes 
accomplished with Orthobase Pro are an “inheritance” from the 
standard aerial photogrammetry. For example, the Z-axis should 
be orientated towards the camera in order to carry out the DTM 
and it is obligatory that the façade be geo-referenced in regard 
to a projection system to fulfil the orthoimage. 
The improvement methods of DSM should include the detection 
of blunders due to spurious correlations in hidden areas. The 
criteria could be purely zone-related before the division of the 
façade into visible/non-visible areas for each stereoscopic pair. 
The detection of other errors may be based on the definition of 
geometric constraints, at least in the façades that may be 
skelotonized into simple surfaces that may serve as a geometric 
reference or context. 
The results improve through a selection of points based on the 
correlation coefficient. The union of the individual DSMs after 
an elimination of gross errors, hidden areas and poor points of 
correlation notably improves the result obtaining a DSM of a 
great quantity of points and a low error quantity. This may be 
the path to follow in order to take advantage of the possibilities 
of the software used. 
The recognition of homologous points in images with different 
perspectives does not seem to be efficiently resolved. A greater 
effort becomes necessary in the design of efficient algorithms 
adap 
com 
musi 
Fina 
dime 
mod 
of 01 
impc 
revis 
phot 
4.6 
Desr 
the 
Uxm 
Inter 
VSM 
Edm 
meas 
Proc 
Com, 
Augi 
Karri 
8. 1 
Phot* 
Phott 
Krau 
Stane 
Li, Z 
accur 
Photi 
1333 
Maso 
photo 
Instit 
Instit 
Otepl 
devel 
Proce 
Corfu 
Patias 
Archi 
Proce 
in syr 
Peuck 
The I 
Digiti 
of Ph 
540. 
Wang 
match 
Sensit 
Wang 
Digita