nd 
ly 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
  
Imput Image 
Histogram 
Contour Image 
  
  
0 5:960 
0 | : "360 
Hough Space Hough Space Histogram 
Figure 5: Intermediate Results for a Cut of an Input Image of 1988 
Table 1: Results of Building Detection 
  
  
  
Number | Evaluated Detection | Branch 
Result Percentage | Factor 
| 1983 75.396 42.9% 
2 1988 86.2% 36.2% 
3 1988 multitemporal 82.7% 36.0% 
4 2001 96.8% 37.9% 
5 2001 multitemporal 97.2% 41.4% 
  
  
  
  
  
4 RESULTS 
To evaluate the multitemporal approach a building detection tool 
for a GIS verification application was assumed. For this task the 
region-based and structural classifier were combined. Only de- 
tected building hypothesis that overlap with the class inhabited in 
the region-based classification result are taken as a building (cp. 
figure 7). The others are unconsidered for the evaluation. 
The evaluation is based on manually segmented buildings in the 
input images of 1983 to 2001 (see manually detected buildings 
in the input image of 2001 in figure 7). Two measurements for 
a detection evaluation described in (Lin and Nevatia, 1998) were 
made: 
100 - TP 
detection percentage = TPT TN (5) 
; 106 - FP 
branch factor = TPT FD (6) 
The two measurements are calculated by making a comparison 
of the manually detected buildings and the automatic results, 
where TP (true positive) is a building detected by both a per- 
son and GEOAIDA , FP (false positive) is a building detected by 
GEOAIDA but not a person, and TN (true negative) is a build- 
ing detected by a person but not by GEOAIDA . A building is 
considered detected if the main part (min 5096) of the building is 
detected; an alternative could be to require that a certain fraction 
of the building is detected. 
5 CONCLUSIONS 
The developed approach shows how temporal knowledge can 
be used in an automatic image interpretation system. Temporal 
knowledge is modeled in a state transition diagram, the probabili- 
ties for state transitions are used as a priori knowledge for a linear 
regression classifier. 
The approach was tested on a dataset containing aerial images 
acquired in 1983, 1988 and 2001. Three object classes are differ- 
entiated: Inhabited area, forest and agriculture. A region-based 
1247 
linear regression classifier uses features like shadiness, unifor- 
mity, contour angularity and straight contour lines to interpret the 
images. 
For the images of 1988 and 2001 temporal knowledge in terms of 
a previous classification and a state transition diagram was used. 
Both images were also processed without temporal knowledge to 
compare the results. 
To evaluate the multitemporal approach a building detection tool 
for a GIS verification application was assumed. The results in 
table 1 show that the proposed multitemporal approach is appli- 
cable. The multitemporal result of 2001 shows less confusion 
between the classes agriculture and forest than the monotemporal 
result. Additional tests are necessary to measure the advantage of 
a multitemporal approach in comparison with a monotemporal. 
The multitemporal approach was tested in the focus of GIS verifi- 
cation, other possible applications are the detection of alteration, 
environmental studies, the development of urban areas and the 
examination of natural disasters. 
REFERENCES 
Bückner, J., Pahl, M., Stahlhut, O. and Liedtke, C.-E., 2002. A 
knowledge-based system for context dependent evaluation of re- 
mote sensing data. In: L. J. V. Gool (ed.), DAGM-Symposium, 
Lecture Notes in Computer Science, Vol. 2449, Springer, Zurich, 
Switzerland. 
Duda, R. O. and Hart, P. E., 1972. Use of the hough transforma- 
tion to detect lines and curves in pictures. Commun. ACM 15(1), 
pp. 11-15. 
Gomez, C., 1998. Engineering and Scientific Computing with 
Scilab. Birkhauser Boston. 
Lin, C. and Nevatia, R., 1998. Building detection and descrip- 
tion from a single intensity image. Computer Vision and Image 
Understanding: CVIU 72(2), pp. 101-121. 
Meyer-Brötz, G. and Schürmann, J., 1970. Methoden der au- 
tomatischen Zeichenerkennung. K. Becker-Berke and R. Her- 
schel, R. Oldenbourg Verlag, München-Wien. 
Müller, S., Weis, M., Liedtke, C.-E. and Pahl, M., 2003. Au- 
tomatic quality surveillance of GIS data with GEOAIDA. In: 
Proc. PIA 2003 - ISPRS Conference on Photogrammetric Image 
Analysis, Munich, Germany. 
Pakzad, K., 2002. Knowledge based multitemporal interpreta- 
tion. In: Proc. ISPRS2002, Graz, Austria. 
Priese, L., Klieber, J., Lakmann, R., Rehrmann, V. and Schian, 
R., 1994. New results on traffic sign recognition. In: Proceedings 
of the Intelligent Vehicles Symposium, pages 249-254. IEEE, 
1994.