the camera and 
; attached to a 
nd the image 
laptop. 
ings using our 
ire 5 and with a 
t matches were 
| clearly see the 
uations. For the 
ly occluding the 
jade generating 
not represented 
tect the overall 
bustness of our 
image in Figure 
e indicating an 
> to the building. 
of the silhouette 
to correct these 
d, on the right the 
nperfections. 
'ction. 
4.3 Accuracy of Orientation 
While the theoretical accuracy of differential GPS is very high, 
there are many practical limitations. This is especially true 
when using GPS in built-up areas. Shadowing from high 
buildings causes poor satellite configurations. At times the 
signal is lost completely. Additionally signal reflections from 
buildings nearby causes so called multipath effects further 
reducing accuracy. As can be seen form Figure 7 on average we 
experienced an accuracy of 7-10 meters in planar coordinates in 
our tests. The Z component of the GPS measurement was 
discarded and substituted by height values from a digital 
elevation map. However the GPS measurements and the 
orientation angels obtained from the compass and tilt sensor 
were sufficiently accurate to serve as approximate values for 
our automated detection. 
Table 1 shows the results corresponding to the images of the 
opera house in Figure 5. In order to assess the accuracy, the 
viewpoint was selected at a position well identifiable in a high 
resolution digital map. The map coordinates thus serve as 
ground truth for our experiments. The planar coordinates of the 
users position are displayed as received from the GPS 
measurement, from the resection after the object recognition 
and from the map. One can clearly see an improvement of 
several meters obtained by our approach. 
  
  
  
  
  
  
  
  
  
X Y Z 
GPS 3513559,82 | 5404697,37 
Resection 3513551,31 | 5404698,50 243,97 
Map 3513550,25 | 5404701,23 245,21 
  
  
Table 1: The position determined by resection has improved 
over the initial GPS coordinates when compared to ground truth 
from the map. 
In Figure 7 the results of four test cases are summarized. There 
is a clear improvement in all cases in the accuracy of the planar 
position. It is interesting to observe that the final accuracy is 
independent of the initial accuracy of the GPS. The final 
accuracy is also not the same for all four buildings. This again 
indicates that the error within the available model data has a 
strong influence on the result. 
Deviation 
(m) 
  
Ed Position received from GPS 
lll Position computed by resection 
Figure 7: Deviation from ground truth in planar coordinates for 
four test cases. The deviation has been reduced by our approach 
in every case. 
S. CONCLUSION 
We have demonstrated the successful implementation of a fully 
automated process for the detection of buildings in terrestrial 
images. The core detection algorithm is based on the 
Generalized Hough Transform. It has proven to be robust 
against clutter and occlusion. The silhouette of a building has 
shown to be a good choice as representation for the shape of the 
building and that compensated the model imperfections of the 
3D city model. The accuracies, which were obtained for the 
orientation, are sufficient for navigation tasks. Within the article 
we have shown the use of 3D building models and close-range 
photogrammetry for location aware applications such as image 
based orientation and telepointing. 
6. REFERENCES 
D. H. Ballard, 1981. Generalizing the Hough transform to 
detect arbitrary shapes. Pattern Recognition, 13(2), pp. 111- 
122. 
E. Baltsavias, A. Grün and L. van Gool, 2001. Automatic 
Extraction of Man-Made Objects From Aerial and Space 
Images (III). A.A. Balkema Publishers. 
Fischler, M. A. & Bolles, R. C. 1981. Random sample 
consensus: A paradigm for model fitting with applications to 
image analysis and automated cartography. Communications 
of the ACM, 24(6), pp. 381-393. 
Fraser, C. S., 1997. Digital camera self-calibration, in: ISPRS 
Journal of Photogrammetry and Remote Sensing, 52, pp. 149— 
159, 
Fritsch, D., Klinec, D., Volz, S., 2000. NEXUS - Positioning 
and Data Management Concepts for Location Aware 
Applications. In: Proc. of the 2nd International Symposium on 
Telegeoprocessing , pp. 171-184. 
Grün, A. & Wang, X., 2001. News from CyberCity-Modeler. 
In: Automatic Extraction of Man-Made Objects from Aerial and 
Space Images (III) , pp. 93-101. 
Hóllerer, T., Feiner, S., Terauchi, T., Rashid, G., Hallaway, D., 
1999. Exploring MARS: developing indoor and outdoor user 
interfaces to a mobile augmented reality system Computers & 
Graphics 23, pp. 779-785 
Hough, P. V. C. 1962, Method and means for recognizing 
complex patterns. U.S. Patent 3,069,654. 
Klinec, D. & Fritsch, D. 2001. NEXUS — Acquisition of Position 
Information for Location Aware Applications using Multi 
Sensors and Mobile Photogrammetry. Proceedings of ION'01, 
Salt Lake City, USA. 
Koenderink, J.J. & van Doorn, A.J., 1997. The Internal 
Representation of Solid Shape with Respect to Vision. In: 
Biological Cybernetics, Vol. 32, pp. 211-216. 
Kraus, K. 1996. Photogrammetrie, 2, Dümmler. 
Ulrich, M., Steger, C., Baumgartner, A., Ebner, H., 2001. Real- 
Time Object Recognition in Digital Images for Industrial 
Applications. In 5th Conference on Optical 3D Measurement 
Techniques, pp. 308-318. 
Wolf, M., 1999. Photogrammetric Data Capture and Calculation 
for 3D City Models. In: Photogrammetric Week '99, pp. 305- 
313. 
—495-