Roessner, Sigrid
coordinates of tie points, camera calibration parameters, ground control information and navigation data which are
explained in more detail in the following:
1. From image matching (3.1) a subset of 3557 regularly distributed tie points were selected and introduced with an a
priori standard deviation of 0.2 pixel.
2. The camera calibration parameters were derived from MOMS-2P geometric laboratory calibration, conducted at
DaimlerChrysler Aerospace Company (Dasa, former MBB), the manufacturer of MOMS-2P. Earlier investigations
showed, that significant deviations from the lab-calibrated parameters and also temporal variations have occurred
(Kornus et al., 2000). Therefore, the camera parameters were introduced as observations with low weights into the
adjustment to allow for self-calibration.
3. From the 59 GPS points, 11 points had to be eliminated due to gross errors, recognized by corrections 3 times
bigger than the root-mean-square (rms) value derived from all points. This is a result of the above mentioned
problems in identification of some point positions in the field (2.2). From the remaining 48 GCP 41 were
introduced with an a priori standard deviation of 10m in planimetry and 5m in height. The rms-error of the final
corrections was 7.9m, 14.3m and 1.6m in X, Y, and Z.
4. From the MOMSNAV navigation dataset only the orbit data were exploited while the attitude data were neglected
due to the problems mentioned above. At the German Space Operations Center (GSOC) of the DLR the orbit
positions are derived by post-processing of the onboard processed GPS data, which are downlinked from the MIR
station. These relatively inaccurate positions are interpolated with high precision orbit models. The resulting orbit
has an internal accuracy of 1-3m and an absolute accuracy of 30-50m (Gill, 1997).
For the evaluated strip, consisting of 28728 image lines, 19 OI were employed. The distance between the OI was set to
1512 image lines, corresponding to 3.7 seconds flight time. Both GCP and orbit positions were previously transformed
into a local cartesian topocentric coordinate system (LTS) based on the WGS-84 ellipsoid, with its origin close to the
center of the evaluated area. As result the estimated parameters of interior and exterior orientation were obtained,
serving as input for the steps described in 3.4 and 3.6.
3.3 Image matching for dense parallax measurements
The Otto-Chau region growing concept in the implementation of Technical University Munich (Heipke and Kornus,
1991) was used to generate a very dense grid in two separate runs of parallax measurements trying to match every pixel
and every fourth pixel. As described earlier, basic matching algorithm is LLSQM. 3 image pairs were used to check for
blunders at this stage. The whole process consists of the following four steps:
computing mass points for backward/forward image pair (bf)
evaluating backward/nadir and forward/nadir matching runs starting from the coordinates given by the bf-results
using the mean of the two resulting nadir coordinate pairs
discard the match if the distance of the mean to the original coordinates exceeds a threshold (here put to 0.5 pixel)
Sn
3.4 Transformation of mass points into object space
Next, the dense parallax measurements were transformed into the LTS coordinate system by forward intersection using
the estimated orientation of the bundle adjustment. Only points imaged by all three looking directions were used. Here,
the object point coordinates are derived from 6 image coordinates by adjustment. If the correction to an image
coordinate exceeds a threshold (here put to 0.5 pixel), the point is discarded. After transforming the LTS into
geographic WGS-84 coordinates, X, Y coordinates were transformed into the state coordinate system of Kyrgyzstan
(Transverse Mercator Projection related to the Krasovsky Ellipsoid, Datum Pulkovo 1942) using the ARC/INFO
software with user-defined datum definition. The ellipsoid heights were converted to orthometric heights using the
world-wide 15 minute gridded geoid heights provided by the National Imagery and Mapping Agency (NIMA, 1999).
3.5 Generation of a regular raster DEM
For interpolation of the regular raster DEM from the transformed mass points the software package LISA (Linder,
1999) was used. Parameterization of the interpolation procedure defines the accuracy of preservation of topographic
elements originally contained in the mass points. The goal was to optimize between morphological detail and
suppression of noise. For the 18km by 12km test area of Maili Suu (Fig. 1) different versions of a regular raster DEM
were derived using a sliding plane as interpolation function. Based on the mass point dataset with 4 pixel minimum
point distance (70m), various DEM with grid point spacing of 50, 75, 100m and median filter sizes of 3x3, 5x5, 7x7
were investigated. For comparison interpolation accuracy of DEM raster generation was determined in calculating
elevation differences between the original height of mass points and their representation in the regular raster DEM and
1262 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.