Geometric correction

Introduction

CHRIS/Proba is a system designed for multi-angular image acquisition of a given target, with the capability along-track and across-track pointing increasing its overpass frequency. In order to in-crease the radiometric signal the platform performs a slow-down manoeuvre consisting in rotating while scanning to keep the target for a longer time under the sensor. Besides, the scan direction is reversed during the acquisition of the second and fourth images to reduce the acceleration needed for the operation. These characteristics introduce strong perspective distortions, especially for the first and last images with larger observation zenith angles.
Many applications require the images to be geometrically referenced and/or rectified. In particular for those that make use of multi-angular information the geometric processing must be very accurate.

Solution

The proposed approach for the geometric correction of CHRIS/PROBA is based on the parametric modelling of the acquisition process. It makes use of the satellite's position, velocity and pointing at the moment of line acquisition, projecting the line of sight onto the Earth surface to calculate the geographical coordinate of each pixel. The coordinates map can then be used for the rectification of the images, or optionally be saved to an Input Geometry (IGM) file for latter use.
Theoretically the algorithm does not require any user input as it is based on the geometry of acquisition. Unfortunately PROBA platform does have a pointing problem, which has not been identified yet. Therefore, at least one GCP per image is required in order to compensate for the de-pointing although the precision of the correction is not good enough. Three GCPs per image provide satisfactory results and nine GCPs per image result in excellent co-registration for the 5 images, for the tested cases (all in Barrax).
The choice of the orbital parametric approach is based on the especial requirements that the peculiar acquisition process of CHRIS/PROBA imposes on the other existing methods for geometric correction:

• The first one is the Ground Control Points (GCP) method, which requires a quite large number of points for an accurate geometric correction (at least 30, and more than 50 for high accuracy over flat terrain), and they need to be evenly distributed. These requirements are for each one of the images in one acquisition set.

• Another possible method is based on the photogrammetric equation applied to the five images but it makes the assumption that each image is acquired instantly from a fixed point in space, which is not CHRIS/PROBA case. In order to compensate this difference in the acquisition assumptions it is necessary the use of GCPs distributed through the image.

The overlap of the five images is not too high, around 65% (estimate) due to de-pointing, but even with perfect pointing, larger observation angles provide different spatial coverage due to perspective. Therefore the co-registration of the five images is only possible in a portion of each reducing the accuracy in the excluded areas. This makes necessary a high-resolution reference image for GCP selection.
Even if the requirements of these methods are met, there is the unsolved problem of open water targets, where it is not possible to use GCP at all, except for very few pixels in the best case. The same situation appears in the case of cloudy images, rendering useless scenes that might have still some useful portions.

• At present the geometric correction prototype has been developed to a working state. It has been tested on the Barrax test site, with Mode 1 images. But further development still is required; especially the use of a DEM for line of sight interception, which has been left for the final implementation stage in order to simplify the model while being developed. Also, a refined GCP input system needs to be included. It needs extensive testing on other sites apart of Barrax, but the corresponding telemetry needs to be provided by ESA as such data is not publicly available yet. It is also necessary to test other operation modes apart of Mode 1, especially the peculiar Mode 5.

• The inputs needed by the algorithm are: telemetry with the satellite position, velocity and image timing, target centre coordinate, local DEM (in the final version). The outputs of this algorithm consist on the coordinates map in IGM files, or the rectified images if the desired output is the rectified images, then they should be also provided. As side product, observation angles maps can be produced for each of the images.

• It must be noted that this method requires ESA to provide the corresponding telemetry for each acquisition. It is possible to use orbit propagation models in order to substitute the telemetry in case it is missing, but it has not been implemented yet.

The use of IGMs might be advantageous for reducing storage space as well as for those processing algorithms that might be computationally intensive but do not required rectified images, as the number of pixels to be processed is greatly reduced. They provide geographic location of each pixel; therefore algorithms based on coordinates can still be applied (if they accept the geographical information in this form).
Spatial correlation is a possible approach to automatically set GCPs to reduce the burden on the user but it still needs supervision on each of the points detected as it is typical to get false matches.

Description of the Acquisition Procedure

The acquisition process has been described in detail in a Technical Note provided by ESA:

Open

The spacecraft is oriented such the instrument line-of-sight is pointing towards the target at all time. This definition leaves one degree of freedom open; the rotation around the Line-of-sight (LOS). PROBA has adopted a convention to fully define the rotation matrix from the orbital frame (roll-pitch-yaw) to the frame defining the attitude of the spacecraft while imaging. Instead of a general sequence of three rotations, only two are used.
First, a pitch rotation is made to bring the pitch-yaw plane onto the target. Then a roll rotation (the new roll) is made to bring the LOS towards the target. This strategy effectively assumes a flat earth. The projection of the instrument slit on the ground is a straight line perpendicular to the ground track when looking straight down towards nadir. It would still be a line perpendicular to the ground track after the two above rotation if the earth were flat. However, the earth sphere effectively distorts the line.

Open

The scanning motion is super-imposed to the above manoeuver by targeting a moving point on the earth instead of targeting always the centre of the image. This point is moving over the area to image in a plane parallel to the orbital plane, effectively rotating back and forth around the orbital axis, "buried" in the earth. In order to keep the same scanning direction for all 5 images, the rotation axis is also frozen shortly (c.f freeze time above) before the beginning of the acquisition and maintained throughout. This compromise keeps the direction of scan.