International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
and seems to get worse with decreasing number of GCPs, at
least for this area with large height differences.
As a next step, we checked the role of the area covered by the
GCPs, using always 5 GCPs (Table 8). RPCI gave more or
less similar results in planimetry, verifying previous
investigations with the 2D affine model. The height however,
is more sensitive to the position of the area covered by the
GCPs, deteriorating in accuracy when GCPs were only in flat
areas. Surprisingly, RPC2 gives clearly worse results than
RPCI, especially when GCPs cover only 1/3 of the image area.
This has been also verified with the Geneva images. A
possible explanation is that after the RPCs are used, the scales
and shears of the affine transformation model very small
residual model errors. If in addition the GCP measurements
are noisy (see e.g. the particularly high RMS at the
mountainous south-west where GCP definition was poor), and
the area covered is small, then these parameters may easily
take wrong values. Grodecki and Dial (2003) mention the need
to use only a linear factor in flight direction if the strip is long
(about > 50 km). In future investigations, we will analyse to
what extent the 4 scale and shear parameters are significant
and determinable. These preliminary results indicate that
RPC2 should be used with a GCP distribution covering most of
the image area.
4. ORTHOIMAGE AND DSM GENERATION
The focus in the following text will be on the DSM generation
in Thun. The results of the orthoimage generation in Geneva
are analysed in Heller and Gut (2004). The accuracy of the
orthoimages generated with the laser DTM and RPC2 with 10
GCPs gave an exceptional accuracy of 0.5 m - 0.80 m for both
IKONOS and QB, with very typical sensor elevation values.
These orthoimages are thus more accurate than the national
Swissimage orthoimages, however interpretation of objects is
more difficult.
4.1 DSM Generation Method
For DSM generation, a hybrid image matching algorithm was
used (for details see Zhang and Gruen, 2003, 2004). Our
method considers the characteristics of the linear array image
data and its imaging geometry. The method can accommodate
images from very high-resolution (3-7 cm) airborne Three-
Line-Scanner images to HRS images like IKONOS, QB and
SPOT-5. It can be used to produce dense, precise and reliable
results for DSM/DTM generation. The final DSMs are
generated by combining the matching results of feature points,
grid points and edges. Matching is performed using cross-
correlation and image pyramids. A TIN-based DSM is
constructed from the matched features (whereby edges are used
as breaklines) at each level of the pyramid, which in turn is
used in the subsequent pyramid level for approximations and
adaptive computation of the matching parameters. The
modified MPGC (Multiphoto Geometrically Constrained
Matching) algorithm (Gruen, 1985; Baltsavias, 1991) is
employed to achieve sub-pixel accuracy for all points matched
(if possible in more than two images) and identify some
-inaccurate and possibly false matches. Finally, a raster DSM
can be interpolated from the original matching results.
Sensor GCP | CP x-RMS | y-RMS | zRMS | max. |max. Ay | max. Az
Model [m] [m] [m] Ax [m] [m] [m]
rpcl 24 - 0.44 0.46 1.06 -1.11 -0.89 2.08
rpc2 24 - 0.39 0.42 0.68 -0.95 -0.84 -1.40
3daff 24 - 237 1.07 0.86 -4.87 2.05 1.57
rpe2 20 4 0.40 0.42 0.68 -1.01 -0.93 -1.41
rpc2 12 12 | 0.41 0.46 0.72 0.90 -0.92 -1.44
rpc2 5 19 1.20.51 0.43 0.90 -1.37 -0.78 -1.40
Table 7. Comparison of sensor models and number of GCPs in the IKONOS triplet (Thun). CP are the check points.
Sensor GCP | CP x-RMS | y-RMS | y-RMS | max. max. max.
Model [m] [m] [m] Ax [m] | Ay [m] | Ay [m]
rpcl 5 19 0.45 0.46 1.10 -1.07 -0.99 2.30
rpc2 -5 19 0.67 1.70 3.45 1.18 -3.04 6.24
rpcl 5 19 0.50 0.47 1.63 -1.33 0.89 2.93
rpc2 5 19 0.82 0.97 1.75 -1.51 2,02 3.17
rpcl 5 19 0.45 0.46 1.25 -1.05 -0.96 2.74
rpc2 3 19 0.53 0.59 1.50 -1.03 -1.52 3.15
rpcl S 19 0.49 0.46 1.65 1.06 -1.05 3.35
rpc2 5 19 0.47 0.86 0.92 -0.95 1.95 1.94
rpcl 5 19 0.45 0.46 1.10 -1.06 -1.16 4.11
rpc2 5 19 0.41 0.70 1.05 -1.18 -1.19 -2.33
Table 8. Different distribution of GCPs in the IKONOS triplet Thun. CP are the check points. In the upper table part the GCPs
cover 1/3 of the image in south-west, south-east, north-east and north-west, respectively (the most mountainous part is south-west,
and then north-east). In the bottom table part, GCPs cover 2/3 of the image.
1254
