Two software programs, marstie and marsnav, work together to generate pointing corrections and rover micro-localization for in-situ images. The programs are based on the PIG (Planetary Image Geometry) library, which handles all mission dependencies. As a result, there is no mission-specific code in either of these programs. This software corrects geometric seams in images as much as possible (some parallax seams are uncorrectable).
First, marstie is used to gather tiepoints. The program analyzes the input image set, determines which images overlap, and presents overlapping pairs to the user. The user then manually creates a number of tiepoints between each pair, by identifying the locations of features that are common to both images. An automatic correlator assists the user in getting subpixel accuracy on these tiepoints. Tiepoints may also be edited.
The tiepoints are then used by the second program, marsnav, to generate pointing corrections. This works by projecting one half of each tiepoint to a surface model and back into the other image. This projected location is then compared to the measured tiepoint and a residual error is determined. A global minimization process adjusts the pointing of each input frame until the optimal pointing is determined. The pointing is typically constrained to match possible physical camera motions, although the pointing model is selectable via the PIG library. The resulting “nav solution” is then input into the mosaic programs, which apply the pointing adjustment in order to make seamless mosaics.
In addition to adjusting the pointing, marsnav can also adjust the surface model (helpful when dealing with an unknown terrain), and the position and/or orientation of the rover itself. The latter results in a
“micro-localization” — determining where the rover is and how it is oriented on a very fine scale.
Commercial mosaic-stitching programs exist. However, they typically perform unconstrained warping of the images in order to achieve a match. This results in an unknown geometry and unacceptable distortion. By correcting the seams using this pointing-correction method, the result is constrained to be physically meaningful, and is accurate enough to be acceptable for use by sci-
ence and ops teams. This method does, however, require a priori camera calibration information. The techniques are not limited to mast-mounted cameras; they have been successfully applied to arm cameras as well.
This work was done by Robert G. Deen and Jean J. Lorre of Caltech for NASA’s Jet Propulsion Laboratory.