|t. 
Ee 
o 
[iieri -ImISIVI 
[erem ET 
  
n in [76]. Overall 
Build.: building; 
'ctness. 
s V M Lesen AN oh, ne itd un NRF BEART» 
IE EIRE ES EI 
0 
43 
02 
— 
  
— 
re given in [%]. 
integrating more 
res related to car 
i-scale features. 
ticated model of 
Cumar & Hebert, 
the association 
Finally, we will 
| by using image 
ence Foundation 
(DFG) under grants HE 1822/25-1 and HI 1289/1-1. The 
Vaihingen data set was provided by the German Society for 
Photogrammetry, Remote Sensing and Geoinformation (DGPF) 
(Cramer, 2010): 
http://Www.ifp.uni-stuttgart.de/dgpf/DKEP-Allg.html. 
  
  
  
  
  
  
  
  
  
  
  
  
Class : d n 2 2, à 
= = = = = $ m = E 
Ref. eila&ailajlsiailslalsale 
Asph. | 25.14 | 1.31 | 1.76 | 0.48 | 0.01 | 0.53 | 0.44 | 0.04 184.62 
Build. 1.80 [13.14] 0.48 | 0.08 - 0.16 | 0.24 | 0.13 | 81.93 
Grass 1.54 | 0.63 | 14.65 | 1.21 | 0.41 | 4.65 | 0.09 | — [6321 
Agr. 0.98 0.31 | 0.75 | 7.32 = 3.231.003 == 158.02 
Beach 0.10 - - - 10.00] -- = - ES 
Tree 0.31 012 | 257 ]-0.18 — |1428| 0.01 — 181.73 
Car 0.26 0.02 | 0.03 | 0.01 - 0.01 | 0.31 - 147.94 
Bridge 0.14 | 0.06 [ -- - - - 10.02 7000| — 
Corr. 83.02 |84.24 | 72.37 | 78.94 | -- |62.48 127.22 | — 
  
  
  
  
  
  
  
  
  
  
  
  
    
    
Table 3: Confusion matrix for the experiment using 8 classes 
and the car confidence feature. All values are given in 
[%]. Overall accuracy: 74.84%. Asph: asphalt. 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Class à = : ; = ; : $ : 
Ref. za ZZ Al ES 
Asph. 24.63 | 1.28 | 1.60 | 0.72 | 0.01 | 0.56 | 0.88 | 0.03 | 82.90 
Build. 1.53 [13.26] 0.44 | 0.11 -- 0.16 | 0.40 | 0.13 | 82.70 
Grass 1.30 | 0.62 | 14.21 | 1.60 | 0.41 | 4.87 | 0.17 - 161.30 
Agr. 0.96 | 0.27 | 0.68 | 7.38 3.25 | 0.07 - 158.53 
Beach 0.10 - - -- 0.00 - m I m 
Tree 0:10 | 0.12 2.36 | 0.27 - [1458] 0.02 - 183.48 
Car 0.25 | 0.02 | 0.03 | 0.02 -- 0.01 | 0.32 -—- | 49.33 
Bridge | 0.12 | 007| - | - | - | - | 003 |000| - 
Corr. 84.95 | 84.79 | 73.49 | 72.99 | -- |62.24]|17.001| - 
  
Table 4: Confusion matrix for the experiment using 8 classes 
without car confidence feature. All values are given in 
[%]. Overall accuracy: 74.39%. Asph: asphalt. 
  
au 
Figure 2: Classification results. Left: original orthophoto; 
centre: ground truth superimposed to the intensity 
mage; right: results achieved with 14 classes and the 
car feature superimposed to the intensity image. 
REFERENCES 
Barsi, A., Heipke, C., 2003. Artificial neural networks fort the 
detection of road junctions in aerial images. In: /APRSIS 
XXXIV (3/W8), pp. 18-21. 
Birchfield, S., Tomasi, C., 1998. A pixel dissimilarity measure 
that is insensitive to image sampling. JEEE-TPAMI 20(4), pp. 
401-406. 
   
   
  
   
    
   
  
   
   
  
    
  
  
  
   
  
  
   
   
  
  
   
     
   
   
   
   
  
  
   
    
    
    
    
   
    
    
  
  
   
   
   
   
Cramer, M., 2010. The DGPF test on digital aerial camera 
evaluation — overview and test design. Photogrammetrie— 
Fernerkundung-Geoinformation 2(2010):73-82. 
Dalal, N. and Triggs, B., 2005. Histograms of Oriented 
Gradients for Human Detection. Proc. of IEEE Conference 
Computer Vision and Pattern Recognition: 886-893. 
Friedman, J., Hastie, T., Tibshirani, R., 2000. Additive logistic 
regression: A statistical view of boosting. Ann. Stat. 28(2):337— 
407. 
Geman, G. and Geman, D., 1984. Stochastic relaxation, Gibbs 
distribution and Bayesian restoration of images. JEEE Trans. on 
Pattern Analysis and Machine Intelligence, 6(6): 721-741. 
Hinz, S., Bamler, R., Stilla, U., 2006. Theme issue: Airborne 
and spaceborne traffic monitoring. ISPRS J. Photogramm. 
Remote Sens. 61(3/4). 
Hirschmiiller, H., 2008. Stereo processing by semiglobal 
matching and mutual information. /EEE TPAMI 30(2):328-341. 
Kumar, S. and Hebert, M., 2006. Discriminative Random 
Fields. Int’l. J. Computer Vision, 68(2): 179-201. 
Kurz, F., Rosenbaum, D., Leitloff, J., Meynberg, O., Reinartz, 
P., 2011. In: Proceedings of International Conference on 
SMPR 2011 (on CD-ROM). 
Leitloff, J., Hinz, S., Stilla, U., 2010. Vehicle extraction from 
very high resolution satellite images of city areas. [EEE Trans. 
on Geoscience and Remote Sensing, 48(7): 2795-2806 
Li, S. Z., 2009. Markov Random Field modeling in image 
analysis. 3" ed., Springer, London, 357 p. 
Lienhart, R., Kuranov, A., Pisarevsky, V., 2003. Empirical 
analysis of detection cascades of boosted classifiers for rapid 
object detection. Pattern Recognition, vol. 2781, pp. 297-304 
Mayer, H., Hinz, S., Bacher, U., and Baltsavias, E., 2006. A test 
of automatic road extraction approaches. In: JAPRSIS XXXVI- 
3, pp. 209-214. 
OpenCV, 2012. 
http://opencv.itseez.com/modules/calib3d/doc/calib3d.html 
Ravanbakhsh, M.; Heipke, C.; Pakzad, K., 2008a. Road 
junction extraction. from high resolution aerial imagery. 
Photogrammetric Record 23 (124):405-423. 
Ravanbakhsh, M.; Pakzad, K.; Heipke, C., 2008b. Automatic 
extraction of traffic islands from aerial images. Photo- 
grammetrie—Fernerkundung-Geoinformation 5(2008):375-384. 
Rutzinger, M., Rottensteiner, F. Pfeifer, N., 2009. A 
comparison of evaluation techniques for building extraction 
from airborne laser scanning. IEEE Journal of Selected Topics 
in Applied Earth Observations and Remote Sensing 2(1):11-20. 
Stilla, U., Michaelsen, E., Sórgel, U., Hinz, S., Ender, J., 2004. 
Airborne monitoring of vehicle activity in urban areas. ZAPRSIS 
XXXV- B3, pp. 973—979. 
Tieu, K., Viola, P., 2004. Boosting image retrieval. Int. J. 
Comput. Vis., vol. 56, no.1-2, pp. 17-36 
Viola, P., Jones, M. J., 2004. Robust real-time face detection. 
Int. J. Comput. Vis., vol. 57, no. 2, pp. 137-154 
Vishwanathan, S., Schraudolph, N. N., Schmidt, M. W., 
Murphy, K. P., 2006. Accelerated training of conditional 
random fields with stochastic gradient methods. 23" Int. Conf. 
on Machine Learning: 969-976.