Correlation errors or objects in the terrain?

The PIL programme gives the option of visualising the deviations between the frame of reference and the generated grid as error arrows. Each landscape type has been assigned a colour. These are:

• Flat terrain = green

• Gravel pit = yellow/orange

• Village = red

• Hilly terrain = violet

• Woods = blue

The error arrows are designed so that negative deviations give a dark downwards pointing arrow, and positive deviations give a light upwards pointing arrow, an example is shown in figure 7.2.

7 The investigation

Figure 7.2 shows that there are few gross errors in the landscape type open, flat terrain and village, whereas there are relatively many unforeseeable gross errors in the landscape type hilly terrain. There are many positive gross errors in the landscape type woods and a few negative ones.

By means of the error arrows generated by the analysis programme and an orthophoto, it is possible to make a closer analysis of where the errors are located, and whether it is a question of a correlation error or of objects in the terrain. This visualisation does not shift the error arrows, as their location after the shift will not fit in with the orthophotos.

The choice has been to take a closer look at two resolutions of images in scale 1.25,000, namely the re-sults for the resolutions 15µm and 60µm . The choice is based on the fact that large gaps have appeared within each resolution in the number of points eliminated, and furthermore, that these two results differ from the tendency in the other images.

Tables 7.18 - 7.20 show that there are many gross errors in the landscape types woods, gravel pit and hilly terrain. The gross errors are symbolised by arrows. The arrows are here scaled up by a factor 20, so that they are easier to find. The result for scale 1:25,000, resolution 15µm and mesh size 25 x 25m is shown in figure 7.3.

Figure 7.2: The error arrows show the deviation between the frame of reference and the generated grid.

Pale colours indicate a positive error, dark colours a negative error. The error arrows are shifted, so that the starting and finishing point for the individual arrows do not adjoin.

Village Gravel pit

Hilly terrain Flat terrain

Woods Village

In figure 7.3 three areas have been extracted, a gravel pit (yellow), a part of the hilly terrain (violet) and a part of the woods (blue). In the chosen gravel pit there are very large error arrows. By far the most of the error arrows are positive, and only a few are negative. Here Match-T has not been able to extend the generation to the bottom of the gravel pit, but has finished by passing it over. In the landscape type hilly terrain there are also many gross errors. After a closer analysis of the error arrows over the orthophoto, it can be seen that a part of the gross errors are due to the farms with outbuildings and vegetation around.

Here the automatic generation has not been able to eliminate these objects so that the grid returned to the terrain. Likewise, a field is seen which, due to the soil conditions, has assumed a funny pattern.

Match-T regards this pattern as shifts in the terrain, but in reality, it is a gently sloping field without any kind of shifts.

The result of the automatic generation is as expected as regards the landscape type woods. It can be seen from the figure that there are many positive errors (large blue arrows), as the generation must have passed over the treetops.

The same areas as in figure 7.3 have been chosen in figure 7.4.

Figure 7.3: The differences between the generated grid in scale 1:25,000, resolution 15 µm and mesh size 25 x 25m and the frame of reference are shown as arrows. Interesting areas have been extracted.

7 The investigation

Figure 7.4 shows that the error arrows generally are much smaller for the whole area.

In the landscape types flat terrain (green), village (red) and hilly terrain (violet), the number of gross errors have fallen compared to the result for images with resolution 15µm in figure 7.3. This can be seen from the small size of the arrows and from the fact that, in some places, there are holes within the landscape type, where the error arrows are too small to appear.

Generally seen, the error arrows for gravel pits and wooded areas are markedly smaller by resolution 60µm.

In the chosen areas, it will be seen that the error arrows are considerably smaller with a resolution of 60µm that with 15µm. As regards the gravel pit, there are markedly fewer of the very large error arrows.

Furthermore, there are both positive and negative error arrows which indicates that the generated grid in Figure 7.4: The differences between the generated grid in scale 1:25,000, resolution 60 µm and mesh

size 25 x 25m and the frame of reference are shown as arrows. Interesting areas have been extracted.

In general, it can be seen from the figures in Appendix K that the error arrows are large when using a high resolution. These error arrows are reduced by using a lower resolution. For images in scale 1:5,000, it is seen that the same area consisting of village and woods include very large error arrows for all the results, regardless of the image resolution or mesh size. A closer analysis shows that the area is covered by a model. Even if the results from this model are poor, they still show that the deviations are reduced by us-ing a lower resolution.