5. ANALYSIS OF RESULTS
The results of testing the anchorpoints technique are
demonstrated for flat; moderate and rough terrain in Figures
1; 3; and 5 respectively. By examining these results one can
conclude that, for all terrain types the eight parameters
transformation within the DEM grid gives more accurate
results compared with affine transformation. This is mainly
because this method insures exact fit of the four DEM points
forming a grid (because there is no redundancy as four points
are used to determine the eight parameters).
The results of pixel by pixel technique are shown in Figures
2; 4; and 6 for flat; moderate and rough terrain respectively.
These Figures show that:
- For flat terrain, the best results are obtained by interpolating
height within the grid by inverse distance method..
- For moderate and rough terrain, the bilinear interpolation
method give the best results.
- For all terrain types, the worst results are obtained by the
nearest neighbour method.
Accordingly one can conclude that a simple formula such as
inverse distance is able to represent flat terrain for
orthoimage production, but more complex formulas are
needed for representing moderate and rough terrain.
In order to be able to compare the anchorpoints and the pixel
by pixel techniques, the method which give better results for
each terrain type is selected to represent the technique. These
methods are shown in Figures 7; 8; and 9 for flat, moderate
and rough terrain respectively.
By examining Figures 7; 8; and 9 one can conclude the
following:
- For flat terrain (Figure 7) it is clear that pixel by pixel
technique gives better results at all grid sizes.
- For moderate and rough terrain (Figures 8 and 9) although it
was expected that the pixel by pixel technique will give
better results, the two techniques give almost the same
results (only a very small improvements, not clear in the
Figure because of its small scale, is obtained at bigger grid
size using pixel by pixel technique). This could be due to that
the used formulas for height interpolation could not represent
such type of terrain accurately and more elaborate algorithms
are needed for this purpose.
Figure (10) demonstrate the influence of DEM density (grid
size) on the accuracy of orthoimages generated for different
terrain types, where the pixel by pixel technique (which give
the best results) is represented in this Figure.
As declared in Figure 10, in all terrain types, the accuracy of
orthoimages decreases by increasing the DEM grid size but
with different rates. The orthoimages of flat terrain are less
affected by the increase in the DEM grid size compared with
the orthoimages of moderate and rough terrain, and the
orthoimages of rough terrain are most affected by the increase
in the DEM grid size. The relationship between DEM grid
size and the accuracy of orthoimages can be represented by
straight lines for all terrain types
| e Series 1
eries 2
1
1 L 1
0 1 1 1
200 300 400 500 600 700 800 900 1000
Grid Size {m}
RMS (of Vector Displacements) {Pixels}
Figure 7. Flat Terrain (Anchorpoints and Pixel by Pixel
Techniques)
Series 1= Anchorpoints 2= Pixel by Pixel
@ Series 1
1 À 1
0 1 1 1 1
200 300 400 500 600 700 800 900 1000
Grid Size {m}
RMS (of Vector Displacement) (Pixels)
nN
Figure 8. Moderate Terrain (Anchorpoints and Pixel by Pixel
Techniques)
Series 1= Anchorpoints 2= Pixel by Pixel
f @ Series 1 A
3 FeFseries 2
0 1 1 1 1 L L 1
200 300 400 500 600 700 800 900 1000
Grid Size (m)
Figure 9. Rough Terrain (Anchorpoints and Pixel by Pixel
Techniques)
Series 1= Anchorpoints 2= Pixel by Pixel
RMS (of VEctor Displacement) (Pixels)
250
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
Th
on
det
ort
pix