Renata Jedryczka
VECTOR DATA IN SEMI-AUTOMATIC CORRECTIONS OF DENSE DEM
FOR ORTHOIMAGES GENERATION.
Renata Jedryczka
Olsztyn University, Poland
Department of Photogrammetry and Remote Sensing
renata.jedryczka Q kfit.uni.olsztyn.pl
Working Group IV/2
KEY WORDS: DTM/DEM, vector data, semi-automation, segmentation, buildings, corrections.
ABSTRACT
Fully automatic generation of dense DEM from aerial images on urban areas is usually not sufficient in a accuracy for
orthoimage creation particularly in the large scales. The correction process of DEM needs a lot of additional manual or
semi-automatic techniques and it is time consuming.
The paper presents the method using additional information from existing vector data (f. e. from databases or from
files including local measurements) to localise and correct DEM areas of buildings and their surroundings.
All calculations were made with the self made software”.
1 INTRODUCTION
The automatic generation of DTM/DEM is in these days the conventional technique used in all digital photogrammetric
workstations. They create files of heights data of various kinds of organisation, usually in regular or irregular grid of
object co-ordinates. When we work with images in large scales (e. g. 1:5000) and particularly with the build-up areas
or with poor texture we need more additional information. (Krzystek , 1998, Ackermann, 1996). It could be taken as
the inner information from the images themselves or as the outer from data bases. The existing maps for example can
serve as the information of localisation of the buildings (Pasko , 1996) or roads (Roux,1996).
In this paper the data in the irregular grid were used because they were the original results of fully automatic DEM
generation from two stereo images on the basis of area based technique. The additional information has been
introduced on the last step of the pyramid of the images. These new data were stored as the vector data. The plane co-
ordinates describing the building were taken into account. So only geometry of the objects was considered. The
proposed method also used the operator as one of the elements of decision what to do next in step by step DEM
calculation process. In the end the dense grid DEM was generated using bilinear interpolation. The grid size was
determined by pixel size of orthoimage.
For accuracy checking the collection of points was made by manual measurement from images using the same self-
soft-system, so only influences of interpolations and introduced outer information can be taken into account. The errors
of Z calculations determined by all orientation processes has been omitted (without changing generality in relative
comparison of heights).
2 SOURCE DATA
The first step was to obtain the data from automatic process of DEM generation from two images. The area based
method with correlation coefficient as the similarity measure was implemented. The all levels of pyramid were used to
create collection of Z data. In the object co-ordinate system they gave the irregular located points.
The calculation process is usually finished when the size of surroundings is bigger than the distance ( in the object co-
ordinate system) between centres of orthoimage’s pixels. If images were taken on build-up areas it can look like in the
Figure 1. There exists a lot of empty parts because correlation coefficient was too small or even not calculated.
" We used programmes made by dr Z. Paszotta: pyramid of images , automatic generation of DEM, determination of
orientation parameters, and author himself; all in Windows NT/95,98 environment.
452 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
