The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008
At first glance the images were cloudless. This impression
resulted from the low contrast between the shadow-projecting
clouds and the snow-covered ground. The older version of LPS
software apparently did not manage to filter excess values.
The other type of errors which came to light with the LPS 9.0
was inaccurate elevation extracted along linear objects oriented
along the left and right edges of the stereopair i.e. parallel (+/- 5
degrees) to the satellite’s orbit (Fig. 3).
Figure 3. Gross errors identification using Blend Tool in
ERDAS Imagine
Both types of errors identified during the experiment resulted
not from the sensor mounted on the satellite, but rather from the
inaccurate sensor modeling by software or its correlation
algorithm improperly designed. Additional confirmation to
these claims is the non-existence of the said errors in the newer
version of LPS (v. 9.2 SP1 or later) based on the same source
data. The new model significantly differs from the previous
ones (Fig. 4).
The differences between models generated by two different
versions of LPS are obvious. With LPS 9.0 areas with gross
errors are apparent while the LPS 9.2 SP1 image is much more
smooth, representing better accuracy of the elevation model.
Correlation analysis of both the old and the new models with
the original stereopair shows major improvement (coefficient of
correlation at 0.85). The LPS 9.2 SP1 image did not have a
single object in the suspicious class which is coherent with the
software’s manufacturer statement, regarding the improved
correlation algorithm for Cartosat-1 introduced with LPS 9.2
SP1.
General Mass Point Quality
%
Excellent % (1-0.85)
87.2
Good % (0.85-0.70)
12.5
Fair % (0.70-0.5)
0.3
Isolated %
0.0
Suspicious %
0.0
Table 2. Correlation quality
7. DSM ACCURACY ASSESMENT
Accuracy of the DSM generated from Cartosat-1 imagery was
measured against the two DEMs. Each model was obtained
independently of the other. After the initial analysis of the DSM
its accuracy was tested against the reference model based on
topographic maps with the scale of 1: 25000. The difference
between the DSM based on the LPS 9.2 SP1 and the reference
model was very narrow. The graph below illustrates the
distribution of the differences, most of which tended to stay
within the +/- 3 m range. That fact proved the high fidelity of
the obtained elevation models.
Figure 5. The reference DEM and the calculated DSM
difference distribution.
The next phase was aimed at verifying the calculated model’s
accuracy at the 6 previously chosen Check Points. The results of
the verification are in the Table 3. Only one of the Check Points
shows noticeable deviation from the others with the value of the
difference above 2 m. In all other cases the differences stay
within the 2 m range, which is further evidence of the high
fidelity of the model.
Figure 4. DSMs calculated in LPS 9.0 (left) and LPS 9.2 SP1.