Supplier: Virtual World Developers Inc. Hawaii, USA
Quality:
Orthocorrection production environment description:
| System name | Geoteket 2.0 |
| System supplier | Spacemetric |
| OS | MS Windows |
| Usage | Geodetic computations and ground control point measurements and storage |
| System name | PKu 2.0 (Precision correction) |
| System supplier | Spacemetric |
| OS | Sun Solaris |
| Usage | Image orthorectification |
| System name | Socet Set 4.2.1 |
| System supplier | Spacemetric |
| OS | Sun Solaris |
| Usage | Framework in which PKu functions are implemented |
Orthocorrection method description:
The satellite scene is ingested, and the scene metadata is interpreted. This information is used to calculate an a priori acquisition model for the scene, including the six Keplerian orbital elements, and three attitude angle offsets.
Control point observations are used to refine the a priori values of the orbit and attitude model parameters. A control point observation consists of a 3-dimensional geodetic coordinate for the ground space, and the corresponding 2-dimensional image coordinate. Corrections to the a priori model parameters are determined by least squares adjustment of the control point observations.
Control point data are digitized from maps, provided by the member countries. The ground control is usually derived from existing topographic maps at the scale 1:25 000.
Number of Ground Control Points, (GCPs), used are a minimum of 16 for each full scene.
GCPs should be well spread throughout the scene. Only GCPs that can be recognized on the satellite image are used, such as road intersections, lakes and islands. Also rivers can be used, if necessery, however stabile GCPs like road intersections are preferred.
In the orthocorrection, parallax errors in spatial positioning caused by oblique viewing are corrected. A Digital Elevation Model (DEM) is used to determine the parallax size. The DEM is first resampled to the image output space, and the terrain displacement vector is then calculated for each output pixel.
In the resampling of the scene, the original image is transformed to the desired frame, pixelsize and map projection, taking into account the acquisition model and parallax corrections. The resampling kernel used is cubic convolution.
In the quality control of the final product the corrected scene is divided into 9 quadrants, and for each quadrant one control point is digitized from maps and compared with the location of the corrected scene. RMS for all 9 points should be less than 25 meters.