Renata Jedryczka
3 DATA CORRECTION
3.1 Main steps
The main idea of the method is to use the vector data to clean elements in DEM file suiting areas covered by buildings.
All calculations have been performed not in the image but in the object co-ordinate system.
First the roofs of the buildings have been automatically detected on the XY plane and only closed contours have been
chosen. For this the algorithm of finding the closed contours has been implemented. It was necessary because of data
type (ascii files with X,Y,Z co-ordinates and codes describing types of objects and points and lines). It has also been
made automatically.
Afterwards the operator indicated the area which was later cleaned on the screen and also appropriate elements in
DEM file were removed. Some objects not found automatically could be detected and cleaned manually by the operator
just clicking on the borders of selected object on the left image. Using projection on the XY plane the areas were
localised. The histogram of the selected area of Z values is accessible so the operator can determine the minimum and
maximum Z values to remove Z values lying outside them from it.
We can’t, of course, forget about discontinuities (introduced manually) and detection large height differences in X and
Y directions (automatically checked gradients). All mentioned above things serve to have as many points as possible as
lying just on the ground.
That way the first step (cleaning ) has been finished and results you can see in the Figure.3.
50 150 250 350 z
Figure 3. The region with the cleaned areas (on the left) and appropriate histogram of Z values (on the right).
The second step is connected with adding new information of Z values (filling). It also uses interactive tools. Namely,
we can add separated points measured straight on the images (with the help of stereoscopic view) and polygons of
fixed Z value. The external data stored in data base tables can also be used now.
Still empty areas are automatically covered by interpolation of weights average from neighbouring points in the
iteration process. Smoothing of the area is controlled by calculating the gradients both in X and Y directions and by
automatic checking the curvatures. After many iterations (the criteria are the fixed gradients and curvatures connected
with the type of surface) we received smooth dense file without holes.
The third step concerns the creation of dense Z value file depending on orthoimage scale. The Z are interpolated by
bilinear interpolation method.
3.2 The accuracy
The absolute accuracy of DEM is closely connected with the accuracy of calculation of all the orientation coefficients
and with the process of automatic generation of the points from stereo images. We were concentrated only on the
influences of cleaning and filling up processes on relative accuracy. Particularly filling can be one of the more
disturbing elements. So we use a set of checking points for permanent control after every new operation. We take
points measured manually on the images using for calculation object co-ordinates the same coefficients as in
automatically generated DEM. In this way we can compare Z values before and after every iterations. The detached (in
comparison with fixed value) values are monitored on the screen so the operator can check the region manually.
454 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.