CMRT09: Object Extraction for 3D City Models, Road Databases and Traffic Monitoring - Concepts, Algorithms, and Evaluation 
the highest frequency of the signal.”). For our situation it means 
that sensor’s geometric resolution determines the object 
resolution, or in other words, the level of detail of the object. 
Table2: Quantitative results. 
scale 
A 
B 
B 
C 
C 
type 
urban 
low 
urban 
low 
urban 
low urban 
low 
urban 
sensors 
airborne 
airborne 
space 
bome 
space 
bome 
space 
bome 
method 
Seeding 
Seeding 
Seeding 
Matching 
Texture 
Analysis 
CFB 
80.5% 
90.2% 
97% 
88.3% 
34.4% 
NFB 
19.5% 
9.8% 
3% 
11.7% 
65.6% 
WFB 
7.3% 
4.8% 
30.3% 
1.7% 
17.2% 
3.2 Qualitative Assessment of Building Extraction 
The qualitative analysis is based on a comparison between the 
building outlines derived by using the proposed automated 
methodology, and manually mapped buildings (image 
restitution). The manual mapping is carried out by a 
professional operator who performs 2D (or 3D) digitization on 
the input data (oriented imagery or orthophotos). 
The residuals of each building comer from the manual mapping 
and the closest point of the automatically extracted shape are 
computed as a quality measure. 
The results are categorized in three groups depending on the 
method used for building extraction (Table 3). The RMS is 
given in pixels. 
Table3: Qualitative results. 
Hough 
Matching 
Texture 
Analysis 
RMS x 
0.937 
0.898 
0.954 
RMS y 
0.914 
0.958 
0.996 
total RMS 
1.309 
1.313 
1.379 
The number of examined objects is the same as in the 
quantitative analysis. 
Note that the figures in Table 3 are based on image residuals. 
They show the difference of the automatically derived comer 
points and the digitized ones in the image. As our data sets were 
acquired with vertical viewing angles these results can be also 
interpreted as planimetric object space residuals. 
But when dealing with images that were captured with oblique 
viewing angles, the buildings must be projected into object 
space in order to carry out a qualitative analysis in the reference 
system. 
4. CONCLUSIONS 
The aim of this work was to propose a method for generating 
DCMs which makes use of images from spacebome or airborne 
line scanning devices, on orthophotos if available and on 
elevation models. Various image processing techniques, such as 
Hough transformation, adaptive region growing, image 
matching, texture analysis, were employed and investigated for 
deriving the strengths and weaknesses of each. A variety of data 
sets were tested, coming from both spacebome and airborne 
acquisition systems. Through the research based on adaptive 
region growing and on the iterative Hough transformation we 
can conclude that the method is very powerful, but has also 
some weaknesses. One is the high dependence on the 
radiometric quality of the input imagery. Furthermore, rather 
small buildings will not be treated correctly. Image matching 
proved to be a very effective, but very time consuming. The 
suggested strategy of texture analysis, although very efficient 
for pattern recognition over areas in small scale imagery, was 
not very successful for extracting individual buildings. 
Through this research partly very good results were obtained, 
but nevertheless further investigations are necessary for 
improving the quality of the results even more. 
Future work will be focused on: 
• Extraction of objects with holes (e.g. houses with 
inner courtyards), i.e. deriving the inner and outer 
boundary of buildings. 
• Research on constraint settings for aggregating 
neighbouring roof parts that belong to one building. 
• Introduction of multispectral information for making 
the algorithms more efficient, especially as far as seed 
point determination is concerned. 
• Extract edges on sub-pixel bases. 
• Integrate a hierarchical approach in order to decrease 
computation time. 
5. REFERENCES 
Gong, P., Marceau, D.J. and Howarth, P.J., 1992. A 
Comparison of Spatial Feature Extraction Algorithms for Land- 
Use Classification with SPOT HRV Data. Remote Sensing of 
Environment, Vol. 40, pp. 137-151. 
Gonzalez, R.C., and Woods, R.E., 1992. Digital Image 
Processing. Reading, MA: Addison Wesley. 
Gonzalez, R.C. and Woods, R.E., 2002. Digital Image 
Processing, Second Edition. Patience Hall, New Jersey, 799 
pages. 
Haralick, R.M., 1979. Statistical and Structural Approaches to 
Texture. Proceedings IEEE, Vol. 67, No.5, pp. 786-804. 
Kraus, K., 1996. Photogrammetrie, Band2, Verfeinerte 
Methoden und Anwendungen. Duemmler Verlag, Bonn, 488 
pages. 
Scholten, F. and Gwinner, K., 2003. Band 12 "Publikationen 
der Deutschen Gesellschaft fur Photogrammetrie, 
Femerkundung und Geoinformation", Bochum, pp. 419-426. 
Vozikis, G., Jansa, J. and Fraser, C., 2003. Alternative Sensor 
Orientation Models for High-Resolution Satellite Imagery. 
Band 12 "Publikationen der Deutschen Gesellschaft fur 
Photogrammetrie, Femerkundung und Geoinformation", 
Bochum , pp. 179- 186. 
Vozikis, G., 2004: Automatic Generation and Updating of 
Digital City Models using High-Resolution Line-Scanning 
Systems. The International Archives of the Photogrammetry, 
Remote Sensing and Spatial Information Sciences, Vol. 34, Part 
XXX. 
Vozikis, G. and Jansa, J., 2008. Advantages and Disadvantages 
of the Hough Transformation in the frame of Automated 
Building Extraction. International Society for Photogrammetry 
and Remote Sensing XXIst Congress, Beijing, China; 07-03- 
2008 - 07-11-2008; in: "Proceedings", Vol. XXXVII. Part B3b 
(2008), ISSN: 1682-1750; 719 - 724. 
Zhang, Y. (2001): A Spectral and Spatial Information 
Integrated Approach for Tree Extraction from High-Resolution 
Digital Imagery. Digital Earth, Fredericton, Canada, 9 pages.