Babak Ameri
manually by an operator, in order to show a realistic quality measure of the process as well. The results are tabulated in
tables (1), and (2) respectively.
Point-IDY 1 2 3 4 5 6 7 8: 9 10: 11 . 12 | 13 , 14 ] RMSE
AX 10.26 0.20 0.09: 0.17: 0.20: 0.10 0.17 0.41 0.16 0.16 0.23 0.12 038 0.18) 022
Ay |002 011 004 016 0.02 0.09 028 0.14 0.15 0.08 001 007 -021 030| 0.15
pre
Az [0.03 0.03 0.12 0.17 005 0.10 0.18 022 003 0.28 0.03 0.26 -0.35 0.17)
62 |044 044 065 065 030 021 061 007 058 023 037 037 037 0.37
o; lo71.070 023.023 035 035 037 054 054 064 042 043 045 0.11
c? |o69 068 016 016 026 0.12 033 020.062 015 030 031 024 0.08
Table 1: Verified coarse building model based on two corresponding aerial images
A comparison of the estimated variances of the model points coordinates with respect to the absolute values of the
differences between the estimated coordinates and the reference coordinates deduce consistency and agreement in both
tests. Moving from the estimation model based on two images toward the one utilizing four images indicates a tendency
in increasing the accuracy of the estimated model points, as it was expected. In addition imposing more images into the
estimation process increase the reliability of the model.
Pop 1 233.5 6 7181921601. 2i MT EE
ax |0f9 020 0.19 0.17 020 0.17 011 028 003 010 023 012 038 0.06] 019
‘ar |-009 016 -0.12.-020 0.10 -017 -027 0.17 0.18 -019 002 -001 0.18 029) 0.17
“az |-015 0.15 0.06 0.23 0.10 008 0.13 0.09 0.14 004 027 023 019 013) 016
s? [as 034 049 049 020 023 064 008.056 023 022 022 030 026
of Jos 030 023 023 023 032 036 030 032 046 02 032 039 0.11
| c? [034034 021 021 026 016 046.017 064 020 025 025 026 009
Table 2: Verified coarse building model based on four corresponding aerial images
The above experimental results show the strength and generality of the proposed FBMV in recovering the reliable and
accurately defined geometric primitives of different sorts of building structures, which is an essential part of any automated
vision system. It shows that the proposed method is capable of working with any complex polyhedral-like object model,
if an appropriate initial hypothesis model is available.
6 CONCLUSION AND FUTURE WORK
The main objective of this paper was to introduce the concept of the FBMV and give some hints of how the information
derived from the model itself can support the verification process.
The problem considered in this study was to determine the precise geometric description of a polyhedral-like building
model given matches between the model primitives and the image features. However, the proposed framework allows
different non-polyhedral object models to be used. A consequent disregarding of this restriction is that the projected
model edges are not necessarily straight edges and the model faces are not inevitably planar surfaces, thus the geometric
routines should be adopted with different geometry. An important aspect of the FBMV method is the ability to solve
the model parameters by simultaneously fitting all the geometric primitives of the 3D model into all the homologous 2D
image features, taking into account the external and internal geometric and topologic properties of the model structure and
imaging process as constraints during parameter estimation. This is important because it allows the earlier initial matches
or the partial matches between the 3D model primitives and 2D image features force the location of other structural
elements of the model. Thereby new matches that can be used to verify or reject the initial estimated model parameters
are generated.
REFERENCES
Ackermann, E., 1984. Digital image correlation: Performance and potential application in photogrammetry. Photogram-
metric Record 11(64), pp. 429—439.
34 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
tog
Br
St
(ed
Fói
Re
Fói
tog
Fói
ma
Fri
Bei
Ind
Gül
ISP
Ha:
met
Lar
and
Let
Vol
Lov
and
and
Siel
Vol:
the
