The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part Bl. Beijing 2008
effect of systematic errors. The distribution of GCPs in the test
field is shown in figure 1.
Figure 1. Distribution of GCPs in the test field
3. DSM AND ORTHOIMAGE GENERATION
A processing system named Pixel Factory, which is developed
by Infoterra France, is used for DSMs and orthoimages
generation. The Pixel Factory system is an integrated system for
aerial and satellite imagery processing. It performs image
matching for all pixels in the images, and can extract dense,
precise and reliable DSMs, and generate orthoimages from
single- or multi-sensor imagery.
The Pixel Factory system has a fully integrated architecture that
manages full processing workflow, resources and data. All the
processing is performed in a fully automated way, including
automated generation of thousands of tie points, automated tie
point filtering to remove blunders, automated global multi
sensor spatial adjustment to refine geometric models, automated
dense DSM extraction, and automated orthoimage rectification
using DSM information and taking into account occlusion. All
image processing are parallelized in order to speed-up
computation. Figure 2 shows the flow chart of Pixel Factory
system for DSM and ortho-image generation.
In this study, all the 15 stereo combinations of the IKONOS and
QuickBird images are processed to generate DSMs and
orthoimages automatically of the test field. A regular grid DSM
with 4m spacing is generated from the raw measurement for
each stereo pair. The computation of orthoimage is made by
projecting the terrain geometry on the raw images using the
DSM. The resolution of generated orthoimages is lm for in
track and cross-track IKONOS stereo pairs, and 0.65m for
cross-track QuickBird stereo and mixed satellite stereo pairs.
Figure 2. Flow chart of DSM and ortho-image generation
4. ACCURACY ASSESSMENT
4.1 Geometric Stereo Model
In conventional photogrammetry, the base-height ratio (B/H),
which is defined as the ratio of the baseline of a stereo pair to
the average flying height above ground level, is used to
represent the stereo acquisition geometry and to predict the
geometric accuracy. However, a major disadvantage is that this
is valid only for stereo models which have no roll. Compared to
standard in-track stereo imaging, high resolution satellite
pointing camera systems provide a variety of stereo geometries.
Thus, a geometric stereo model is necessary to analyze the
relationship between the stereo acquisition geometry and the
geometric accuracy in a variable stereo model geometric
environment (Cain, 1989).
1146