International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B6. Istanbul 2004 
that we want to extract (maximum and minimum area, 
minimum height, etc.). From the original digital surface model 
it is possible to obtain a digital terrain model through an 
opening process (Baltsavias et al., 1995) that eliminate the 
constructions or another obstacles that can be present on the 
terrain. The difference between both models, DSM and DTM, 
will allow obtaining a new model that contains only the objects 
located on the terrain (buildings, trees, etc.). Using this new 
model, a binarization process is applied. The binary model will 
contain the regions in which the buildings can be located. These 
regions will be examined using shape descriptors as the 
invariant geometric moments (Gerke et al., 2001) obtaining for 
each one of the extracted regions the mass centroid, the 
surrounding and minimum bounding box as well as the height 
of the region (Figure 11). 
| DSM (IMAGE) 
| -— mathematical morphology 
ai 
DSM -* DIM 
DSM-DTM 
— Binarization 
Labeled of regions 
REGIONS | 
| — Moments 
BOUNDING BOX 
  
  
Figure 10. Location step 
IMAGE DTM 
DSM-DTM 
  
BINARIZATION MOMENTS 
   
Figure 11. Partial results obtained from the location 
(building detection) 
These zones that are selected from the terrain are the input data 
for the following steps that provide the characteristics of the 
extracted buildings (Figure 12). 
It is possible to project the bounding box (that probably 
includes a building) from ground —terrain- space into image 
space by means of orientation parameters. Then, we can apply 
different algorithms in this region to obtain the most 
representative rectangle of the building. 
172 
The analysis applied to these images has different steps. First, if 
the original image has a low contrast, a linear histogram stretch 
is applied obtaining an optimum contrast balance in the image. 
Then, in order to eliminate minor significant edges and noise 
that could be presented in the image, an image smoothing by 
means of a gaussian filter is carried out. 
Next step, it is a Canny filtering in order to obtain the image 
edge. Since the filter is applied in a local level (that contains the 
extracted built area) we obtain two main gradient directions that 
correspond with the main building orientations. Then, we use 
the Hough transform for the straight lines extraction 
considering these orientations and an additional condition about 
the number of pixels involved (Figure 13). 
Once the principal straight lines of the considered image region 
are extracted, the possible intersections among them are 
calculated. All the possible rectangles considering the 
intersection points are generated. Different weights will be 
assigned to these rectangles. These weights will depend on the 
number of edge pixels in the rectangle and its similarity with 
the bounding box of the corresponding region. A final rectangle 
is defined, this rectangle is combined with the information 
derived from the DSM in order to obtain the height of the 
building and to create its prismatic model (Figure 14). 
BOUNDING BOX 
+ 
Z Region | 
| 
IMAGE-WINDOW | 
Noise's elimination 
edges extraction (Canny) 
^ Hough-Straight lines 
Rectangles algorithm 
RECTANGLE 
— Z Region 
|! RECTANGULAR PRISM | 
| (Building Representation) | 
Figure 12. Building Extraction 
Using the BUILDING program, the students will be able to 
know one of the most complex procedures related with the 
photogrammetric processes automation. The program provides 
an entire set of all the partial results and finally, its results are 
obtained in a 3D AutoCAD compatible format. 
Using this final file, the results can be evaluated and edited 
using a conventional stereo digital photogrammetric system. 
This task will allow to the students analyze the own limitations 
of the methodology, and learn to necessity of the edition 
processes in the automatic information extraction from 
photogrammetric imagery. 
4. CONCLUSIONS 
The Hough Transform is a very powerful and simple tool for 
the automation of different photogrammetric process. Using 
OIA and BUILDING programs, the students can obtain (and 
analysis) all the partial results of the different steps in both 
processes (automatic inner orientation and prismatic building 
models extraction).