7000 
  
  
  
  
  
  
  
  
  
  
  
  
6000_ 
5000 
Z 40001 
© 3 
= 
rat 
© 3000 
e 
v. 
2000... 
0J BE . — : 
cow oL noe m UT 
ES e f S abntunst 
o Cs bs hb 5: bor 
> Un e Un 9 e 
Height Differences in Metres 
Fig. 3 Histogram of Height Differences 
Band 2 (Green) - Band 1 (Red) 
= 
= 
© 
& 
Le 
© 
= 
Z 
  
0'T<S0 BE B 
soie | | | | 
e 
e 
V 
e 
Un 
0°7- 1opun 
sı-<oz-B C : | | 
0'1-<S'I- 
S'0- «0'I- 
0c«SsI 
0° 240qŸ 8 i ; e |: 
Height Difference in Metres 
Fig. 4 Histogram of Height Differences 
Stereopair - Block (Urban) 
It is obvious from Figure 2 that Band 3 (Blue) is 
significantly different from Band 1 (Red). Similar statistics 
are produced when comparing Bands 3 and 2 (Green). In 
both cases the Blue Band generation was formed 
noticeably higher than the Red and Green. Band 1 and 
Band 2 appear to be very similar to each other (Figure 3). 
This was confirmed in early comparisons with a semi- 
analytical generation. In this case they both displayed 
similar statistics as compared to the semi-analytical grid. 
Comparing with the semi-analytical grid, the Blue 
generation is significantly above, noticeably worse than 
the Red and Green. Continued testing will produce results 
without the influence of these unwanted DEM points and 
will attempt to ascertain in which terrain characteristics 
each band performs the best. 
6.1.2 Urban 
With the urban imagery, two situations will be presented 
here. Firstly, the Block/Stereopair problem is tackled 
again and secondly Minimum and Maximum Template 
Sizes are described. 
With the Block and Stereopair comparison the statistics 
are perhaps more representative than the corresponding 
comparison with the rural imagery, since all the points 
generated were within the predefined boundary and were 
all relevant to the comparison. The magnitude of height 
differences is illustrated in Figure 4. 
In this case 2.9% lay outwith + 2m of each other, but it 
must be remembered that this imagery was at a scale of 
1:3000 and should therefore be within a smaller tolerance 
than the rural 1:10000 imagery. Approximately 8% of 
points are outside + 1m, which is a more suitable level to 
consider. This is a significant difference when considering 
the parameter. 
Minimum and Maximum Template Sizes refer to the 
dimensions of the correlation template, which is square in 
shape. For each point to be correlated, the algorithm 
starts the calculation with the Minimum Template Size. If 
the correlation is unsuccessful, the template size is 
increased. Further failures mean that the size is 
increased up to the Maximum Template Size. If this to 
proves unsuccessful, the height of the point in question is 
interpolated from neighbouring points. 
Figure 5 shows the comparison between the default 
settings and the next setting up i.e. Minimum and 
Maximum Template Size of 7 and 9 compared with 7 
and 11. This appears to be quite a small parameter 
change, considering the limits are 5 and 20 respectively, 
but it does have a considerable effect. 10.596 of points lie 
outside + 1m. of the other surface. 
6.2 ImageStation 
6.2.1 Rural 
Similar to the ERDAS approach, the MATCH-T correlation 
may be performed on any one of the three bands of 
colour imagery. Comparisons between all three of the 
generations provided similar conclusions to the 
corresponding ERDAS comparisons. Selected statistics 
are shown in Figures 6 and 7. Again, Red and Green are 
very similar and Blue is significantly poorer - 
predominantly above Red and Green. 
Within ImageStation MATCH-T the software has been 
designed to make the operators task of parameter 
selection much more simplistic. Here, the operator is 
required to select a Terrain Type, which best describes 
the terrain present in the area to be modelled. The 
options available are Flat, Hilly or Mountainous. On 
selection of one of these three, two parameters are 
influenced by the decision, namely Parallax Bound and 
Epipolar Line Distance. These govern how much above 
and below the predicted position of the point the algorithm 
will search and along how many of the epipolar lines of 
the stereo image the algorithm will operate. 
360 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996 
Th 
im: 
de 
inc 
se: 
Co 
anc 
anc 
lie 
ime 
exp 
diffi 
Cor 
rem 
app 
ana 
6.2. 
Due 
test 
rest 
this 
Cle: 
the 
cons 
the 
wha 
have 
from 
sign 
signi 
Inve: 
beer 
benc 
SUCC 
para