12
e German
, per pixel
on for the
D surface
d to form
nly local,
e size and
le exterior
age strips
netric and
mbination
workflow
is ongoing
increasing
capability
ble blocks.
f 21 HRSC
block
idle adjust-
single orbit
sion of this
including à
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B4, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
large block comprising 21 strips are presented in chapter 4. The
paper concludes with some observations for future work.
2 SINGLE ORBIT STRIPS
The workflow for a systematic bundle adjustment of the HRSC
image data can be divided into four parts: image pre-processing,
tie point matching, bundle adjustment and evaluation. In addi-
tion to the HRSC images, the nominal exterior orientation and
the Mars Orbiter Laser Altimeter (MOLA) DTM are used as in-
put. The adjusted exterior orientation is the output (see Fig. 2).
The use of standard parameters for each processing step allows to
apply this workflow to the entire HRSC data set without manual
intervention. The most relevant parameters are described in the
following sections.
I HRSC images ] / ( nominal orientation data |. T
MOLA DTM
V
image pre-processing
rectified, filtered images
tie point matching
tie point coordinates
bundle adjustment
adjusted orientation data
evaluation
Figure 2: Workflow for systematic bundle adjustment of HRSC
image data
2. Image pre-processing
The first step for the processing of single orbit strips is image
pre-processing. In this step only the panchromatic channels of
the HRSC images are used. Initially a low pass filter is applied
to reduce image noise. A simple Gaussian with a 3x3 pixel filter
matrix have shown good results and leads to a more robust tie
point matching (Schmidt, 2008).
Subsequently the nominal orientation data and the MOLA DTM
are used for a pre-rectification of the image data (Scholten et al.,
2005). This step compensates scale differences and corrects non-
quadratic pixels.
22 Tiepoint matching
In the nadir image the tie points are arranged as a regular grid.
Given the desired number of points (approx. 45.000) and the im-
age size a corresponding grid width is calculated. In contrast to an
interest operator, the grid ensures a more regular distribution of
the tie points. Tie point matching is carried out in the pre-rectified
images. Depending on the image quality, the final number of suc-
cessfully matched tie points lies between 20.000 and 45.000.
Each candidate point in the nadir channel image is matched with
the other four panchromatic channels. Because of higher macro
pixel size and consequently a lower resolution, the color channels
are normally not used for photogrammetric processing.
For image matching a pyramidal approach is used to take large
parallaxes and imprecise nominal orientation values into account.
As similarity measure a normalized cross correlation coefficient
is employed. A 35x35 pixel correlation window is used. Subse-
quently the matching results are optimized by multi-image least
squares matching (Schmidt et al., 2008).
2.3 Bundle adjustment
In the next part of the processing chain a combined bundle ad-
justment is carried out using the image coordinates of the tie
points, the MOLA DTM and the nominal exterior orientation val-
ues (Spiegel, 2007a).
The used bundle adjustment method is based on the well known
approach used in photogrammetry, by which the exterior orien-
tation is simultaneously determined for all images. In case of
HRSC the images are not acquired from individual view points
but in a continuous motion. Thus, the exterior orientation must
be modeled along the spacecrafts trajectory as a function of time.
Three decades ago the concept of orientation points was proposed
to solve this problem (Hofmann et al., 1982). Today, it is a com-
mon approach in the processing of multi-line sensor data.
To fit the photogrammetrically determined 3D coordinates of a
photogrammetric bundle block to a regional or global reference
system, an adequate number of ground control points is normally
used in aerial or spaceborne photogrammetry. For areas with-
out ground control points available DTMs can be used to obtain
an absolute fit (Strunz, 1993). On Mars the MOLA DTM pro-
vides the best global accuracy. Therefore, a combined bundle
adjustment for HRSC image data and the MOLA DTM as con-
trol information was developed, implemented and tested (Spiegel,
2007b).
The general approach of the bundle adjustment is a nonlinear
least-squares adjustment. This optimization aims to find the best
set of unknown model parameters to explain the observations. For
a combined adjustment of HRSC images and the MOLA DTM
there are four types of observations: image coordinates, orienta-
tion parameters, unknown bias and drift parameters to compen-
sate systematic effects and DTM information. The four types
of observation equations used in the bundle adjustment are de-
scribed in (Bostelmann and Heipke, 201 1).
The stochastic model of a bundle adjustment can be modified by
changing the a priori standard deviation of a particular type of
observation. This allows to handle observations as constants with
oo = 0, as free unknowns with c9 — oo or as stochastic values
with a oo of something in between. The strategy specially devel-
oped for HRSC data divides the bundle adjustment into two parts
with different values for the stochastic model (Spiegel, 2007b).
The image coordinates of the tie points are always introduced
with an a priori standard deviation of 00,2y = 1um.
In the first part only the nominal orientation parameters at the ori-
entation points and the tie points are introduced, but no drift and
bias and no DTM information. The a priori standard deviation
for the orientation is chosen as oo,jwx = 0.028gon for the atti-
tude and 00 xyz = 0.01m for the position. The latter considers
the stable orbit of the spacecraft. This part optimizes the internal
accuracy of the strip.
In the second part observations for bias and drift are introduced
with 00 bias, xyz = 1000m allowing a translation of the whole
strip. À drift for the height component of the exterior orientation
is permitted by setting 00 drift,z = 0.01m/imageline. For X
and Y no drift is introduced. For the orientation angles neither
285
