| the query 
metric to 
oceed with 
In order to 
gment the 
ctors, each 
nage. This 
igurations. 
roceed by 
with the 
500), (900- 
100-3000), 
ion we can 
wt of one 
v using a 
ly captures 
ch for the 
' objects is 
image we 
ew objects 
lis sector is 
in the real 
bjects and 
in the real 
rations that 
95% in our 
orientation 
) produce a 
istance and 
| metric we 
h the one 
position the 
orithms we 
odel of the 
ments were 
M and the 
.0. In this 
riments, to 
ition and a 
rresponding 
in different 
es shown in 
their initial 
iorama is at 
rersa. In the 
25 different 
norama was 
ribed in 3.1 
ects in each 
tified seven 
6 right) we 
e matching 
6,7,8, and 9 
the bottom 
bregions are 
3,4,5,6,7,8,9, and 10 for the top panorama, and 1,2,3,4,9, and 10 
for the second panorama. 
  
12345678 910 
Top 445767911105 
Bottom91195133 4 5 8 
  
  
  
Figure 4: Number of objects in each section in the two 
panorama images in figure 5 
When comparing the first image and the top panorama we have 
to consider 488 different configurations (of 7 objects each), 
while we have to consider 410 different configurations for the 
second panorama image. After applying both distance and 
orientation similarity metrics with a weight of 0.75 and 0.25 
respectively, the algorithm successfully identified the right 
configuration, highlighted in the corresponding panorama. The 
orientation angle was retrieved with an accuracy of 1 degree. 
The total time required to process the distance similarity metric 
for the first image against both panoramic images was 56 
seconds. An additional 0.64 seconds were required for the 
computation of the orientation metric for the 35 highest 
candidates that survived the distance comparison. Thus, the total 
time required to recover the orientation of image 1 was 
approximately 57 seconds. 
For the second image we have 890 different configurations (of 5 
objects each) for the first panorama and 772 for the second 
image. In this case the algorithm retrieves successfully the right 
configuration when compared to the second panorama image, 
while the right configuration was ranked 5™ among the 
candidate configurations of the first panorama image. We can 
see that by having a scene comprised by fewer objects we run 
the risk of having numerous potential matches, increasing the 
chances that we might have random configurations that 
resemble closely our scene. However, configurations of fewer 
points produce substantial gains in time requirements. For this 
experiment the computation of the distance similarity metric for 
all the possible combinations in both synthetic panoramas was 
44 seconds, and an additional 0.1 second was required for the 
computation of the orientation metric for the 35 highest 
candidates. The accuracy of azimuth recovery was 
approximately 2.5 degrees for this set-up. 
5. COMMENTS 
In this paper we presented a novel hierarchical approach to 
recover rapidly the orientation of motion imagery in urban 
areas. We proceed by comparing object configurations depicted 
in this imagery to potential configurations that can be formed 
using the virtual database. By considering abstract properties 
such as distances between objects and their relative positions, 
we produce a fast algorithm that supports the rapid recovery of 
image azimuths with accuracies on the order of 1-2 degrees. 
Considering that the associated computational time 
requirements remain below 1 minute per frame of incoming 
image, the presented approach is highly suitable for the dynamic 
nature of motion imagery applications. This is an essential step 
to support the use of motion imagery for the detection of 
changes and the subsequent updating of virtual models of large 
urban scenes. 
REFERENCES 
Agouris P., A. Stefanidis, and S. Gyftakis, 2001. Differential 
Snakes for Change Detection in Road Segments, 
Photogrammetric Engineering & Remote Sensing, 67(12), pp. 
1391-1399. 
Agouris P., K. Beard, G. Mountrakis, and A. Stefanidis, 2000. 
Capturing and Modeling Geographic Object Change: A 
SpatioTemporal Gazetteer Framework, Photogrammetric 
Engineering & Remote Sensing, 66(10), pp. 1241-1250. 
Antone M. and S. Teller, 2000. Automatic Recovery of Relative 
Camera Rotations for Urban Scenes, Proceedings of CVPR, 
2000, Volume II, pp. 282-289. 
Blaser A., 2000. Sketching Spatial Queries. Ph.D. Dissertation 
Thesis, University of Maine, Orono, ME. 
Brenner C., 2000. Towards Fully Automatic Generation of City 
Models. International Archives of Photogrammetry & Remote 
Sensing, Vol. 33, Part B3/1, Amsterdam, pp. 85-92. 
Coorg S. and S. Teller, 1999. Extracting Textured Vertical 
Facades from Controlled Close-Range Imagery, Proceedings of 
CVPR, 1999, pp. 625-632. 
Day A., V. Bourdakis, and J. Robson, 1996. Living with a 
Virtual City, Architectural Research Quarterly, Vol. 2, pp. 84- 
9]. 
Egenhofer M. and R. Franzosa, 1991. Point-Set Topological 
Spatial Relations. International Journal of Geographical 
Information Systems, 5(2): 161-174. 
Goyal R. and Egenhofer, M., 2002. Cardinal Directions 
Between Extended Spatial Objects. IEEE Transactions on 
Knowledge and Data Engineering (in press). 
Gruen A., M. Sining, and H. Dan, 1996. 3-D City models for 
CAAD-supported analysis and design of urban areas. /SPRS J. 
of Photogrammetry and Remote Sensing, 51 (4), pp. 196-208. 
Gruen A., and X. Wang 1999. CyberCity Modeler, a tool for 
interactive 3D city model generation. Photogrammetric 
Week'99, D. Fritsch and R. Spiller (Eds.), Wichmann Verlag, 
Heidelberg, pp. 317-327. 
Haala N. and C. Brenner, 1999. Virtual City Models from Laser 
Altimeter and 2D Map Data, Photogrammetric Engineering & 
Remote Sensing, 65(7), 787—795. 
Jepson W., R. Liggett, and S. Friedman, 1996. Virtual Modeling 
of Urban Environments, Presence, 5(1). 
Kirby S., R. Flint, H. Murakami, and E. Bamford, 1997. The 
Changing Role of GIS in Urban Planning: The Adelaide Model 
Case Study, International Journal for Geomatics, 11(8), pp. 6- 
8. 
Lindeberg T., 1994. Scale-Space Theory in Computer Vision, 
Kluwer Publishers. 
Ranziger M. and G. Gleixner, 1997. GIS-Datasets for 3D Urban 
Planning, Computers, Environments & Urban Systems, 21(2), 
pp. 159-173. 
Stefanidis A., P. Agouris, M. Bertoloto, J. Carswell, and H. 
Georgiadis, 2002b. Scale- and Orientation-Invariant Scene 
Similarity Metrics for Image Queries, International Journal of 
Geographical Information Science, 16(8), in press. 
—165—