Figure 6 — Examples of images of block “fountain” (on the left)
and “cloister” (on the right)
CONCLUSIONS AND PROSPECTS
On the basis of the test blocks processed up today, we are
confident that the algorithms we implemented and the whole
orientation procedure are sound. The applications we had in
mind in the first place, when starting to work on this project,
where mainly in architectural photogrammetry, but there is no
need to stick to this field: for instance, we successfully used it to
orient images taken for camera calibration, with strong
convergent geometry as well as additional shots with +90°
rotations around the camera axis.
There are still cases where the algorithm fails to find a solution
(and therefore the image cannot be added to the block), even
after the procedure went into all steps so all available tie points
in the image have been used. This was for instance the case with
the block “fountain”, where tie and control points were both
signalized. We had available two sets of image coordinates: one
measured at the analytical plotter, the other with template Ls.
matching. This second version had less tie points available,
since l.s. matching could not manage to correlate successfully
some targets, due to strong perspective distortions, while the
operator at the plotter could. Therefore, the procedure run
smoothly with the manually measured block and stopped with
the other. This suggests that, irrespective of whether this
happened because the points at hand in that particular image
were close to a singularity configuration or not, it is worth
taking a users' approach: just measure additional tie points until
the problem is solved. As underlined in the introductory section,
there is no reason for the procedure to fail if block geometry is
sound: in our view, provided that the camera stations are well
placed, it is only necessary to improve locally the connections
of an image to the block; if that does not help, then more images
are necessary.
Future developments therefore will have to go towards the
integration of the procedure in a measurement environment, to
be able to spot the problem and solve it quickly. This is not the
key point, anyway; we want also to improve the robustness of
the algorithm and the self-diagnosis provided to the user about
block geometry and measurement errors.
At the time of writing, we have embarked in an analysis of the
sensitivity of the method to cope with measurement errors.
Though only very preliminary results can be anticipated, they
are useful hints at what should be done. When introducing large
errors (up to 400 um) in the image coordinates of a GCP of the
initial kernel, if only four points are available there is no chance
to get the image oriented. If more points are available, the
image is oriented at a later stage, when redundancy from the
block adjustment is larger and the error shows up in the
residuals. If the error is very large, though, the threshold on the
discrepancy between the given GCP coordinates and those
computed by intersection may be exceeded. This is safe When
redundancy is low and embarking wrong measurements is
dangerous, unnecessary later, when redundancy is enough. This
suggests that we may have to revise some thresholds values or
make them more flexible, according to the work progress.
References from Journals
Albertella, A., Scaioni, M., 1999. Orientamento diretto in close-
range photogrammetry. In Atti della 3a Conferenza Nazionale
delle Associazioni Scientifiche per le Informazioni Territoriali e
Ambientali, Napoli, Italy, pp. 59-64.
Crespi, M., Marana, B, 1995. Un programma per la soluzione
del problema di space resection in configurazione qualunque.
Bollettino della SIFET, no. 1, pp. 59-86.
Fischler, M.A., Bolles, R.C., 1981. Random Sample Consensus:
a Paradigm for Model Fitting with Applications to Image
Analysis and Automated Cartography. Comm. ACM Vol. 24,
no. 6, pp. 381-395.
Killian, K., 1955. Über das Rückwärtseinschneiden in Raum.
Ósterreische Zeitschrift fuer Vermessungwesen, 4, 97-104 and
5, pp.171-179.
Lohse. P., Grafarend, W., Schaffrin, B., 1989. Three-
dimensional point determination by means of combined
resection and intersection. Conference on 3D measurement
techniques, Wichmann, Wien, Austria, pp 1-17.
Sansó, F., 1973. An exact solution of the roto-translation
problem. Photogrammetria, no. 29, pp. 203-216.
Tan, Z., Brandstátter, G., Xu, X., 1996. A method for solving
the inverse problem for photogrammetry. Int. Archives of
Photogrammetry and Remote Sensing Vol. XXXI, Part B3, pp.
806-811.
Zeng, Z., Wang, X., 1992. A General Solution of a Closed-
Form Space Resection. PE&RS Vol. 58, no. 3, pp. 327-338.
References from Books
Cooper, M.A.R., Robson, 1996. Theory of Close Range
Photogrammetry. In Close Range Photogrammetry and Machine
Vision, S. Atkinson K.B. (ed.), Whittles Publishing, Caithness,
Scotland, UK: pp. 9-51.
Wrobel, B.P., 2001. Minimum Solution for Orientation. In
Grün, Huang (Ed.s), *Calibration and Orientation of Cameras in
Computer Vision", Springer-Verlag, Berlin Heidelberg, pp. 7-
62.
Aknowledgements
The authors are indebted to Mattia Crespi for providing the
original version of RESECT. They are also grateful the Prior of
the San Giovanni Monastery in Parma, who allowed to survey
the cloister, used as one of the tests. Thanks also go to Fabio
Dalla Casa, student at the Parma University, which cooperated
to the survey and orientation of the test blocks. Last but not
least, we would like to aknowledge the contribution of Dr.
Alberta Albertella in testing the RESECT algorithm. This work
has been partially funded under the national program
COFIN2000.
7458
“A
KEY W
ABSTR
During t
caused ¢
extraord
for exca
real nee(
excavati
the uses
we have
Finally s
in north
produce
1. INTR
A major
an accul
situation
drawn ir
the lowe
the situa
program
monume
to recon
anything
drawing
cases w
purposes
invaluab
This bec
all over
and the
Traditioi
in a bet
classic 1
automat
sometim
give var
plans. E:
need hi
surveyin
scale wi
years t]
moderni
similar c
photogre
