Science Directed Spacecraft Autonomy
for Mars Missions

Objectives

The rapidly growing field of remote sensing is beginning to supply massive amounts of high resolution imagery of the earth and other planets. Within the earth sciences, parallel supercomputers have always played a prominent role in the visualization of this imagery.

One obvious process of great physical interest is the motion of surface faults during the course of an earthquake. Modern satellites generate images containing millions of pixels, but due to enormous computational demands, workstations are capable of providing ground displacements for only a few hundred pixel locations scattered throughout an image. We would really like to make a high resolution ground-displacement ''map'', showing the movement of every single pixel during an earthquake. This same system can be used to analyze any images where small surface motions are suspected.

Researchers at NASA are now turning their attention to the implementation of powerful new analysis tools for satellite imagery on MPP's, in order to obtain direct information about important physical processes that until now has often remained buried within these large datasets.

Approach

The system could be deployed onboard a spacecraft to select regions of interest to be imaged in higher resolution. Given a library of reference images from previous missions, future missions could apply the system in real time to search for regions of possible surface motion since the reference images were taken. The software could then flag regions that show movement for retargeting at high resolution as high-payoff scientific targets.

Motion Mapping and Mars

    Since we have first visited Mars, images returned have always revealed a sandy, desert like environment. On Earth, dunes move with the wind and this movement can be detected from orbit using the motion mapping software that we have developed. It is still an open question as to whether the dunes on Mars are active or have been still since Mars lost most of its atmosphere. We have been applying the motion mapping software to images from Mariner 9, Viking 1 and 2, and Mars Global Surveyor in order to try to detect active dunes on Mars.

The Motion Mapping Software System

    The ground based, MPP implementation of our system draws from previous work on QuakeFinder. It is executed from a graphical user interface (GUI) in Matlab which processes the data before it is fed to a suite of parallel motion mapping software and then displays the results graphically. The parallel software is scalable and can be run on a single machine or an entire farm of workstations when the analysis task is large. This system allows for easy experimental setup and fast results so that one can explore the data effectively without waiting for long runs or interpreting difficult to read output files.

Accomplishments

Our First Experiments

    We began our investigation using images of the Proctor Dune Field in the Hellisponte Montes region on Mars. The Mariner imagery was scanned from photographs (194b5.img, 229b5.img, and 229b6.img). The Viking imagery was downloaded in JPG format from the JPL Planetary Image Atlas (http://www-pdsimage.jpl.nasa.gov/cgi-bin/msearch.pl) at 50 south, 330 west.

    The first image pair analyzed was m47s330w-01.pgm (from 229b5.img) and v47s330w-01.pgm (from viking-hellisponte.jpg). In order to obtain this pairing, the Mariner image was lightened, and the Viking image was substantially enlarged (see Figure 1 to compare the original images). These images were then compared using QuakeFinder in the following experiments:

Experiment #1
QuakeFinder tile controls
tile-size 50 50
tile-skip 10 10
max-disp 20
Experiment #2
QuakeFinder tile controls
tile-size 100 100
tile-skip 10 10
max-disp 5

Figure 2: The vector field show the results of these experiments.

    As you can see in Figure 2, there is no overall pattern apparent in the vector fields. While there are some small regions of similarly pointed vectors, these regions are easily explained by examining the correlation surfaces associated with each tile. The set of similarly pointing vectors usually represent the jump form one dune top to the next or a jump to a place along the the top of the correct dune but in the wrong location. This can be seen in Figure 3, an example of jumping to the next dune top, and in Figure 4, an example of moving along the correct dune top. While these results could indicate that there has been no dune motion, recent images from MGS strongly indicate that there are active dunes on Mars.

A Second Set of Experiments

    A limiting factor in the above experiments was the fact that the Mariner image used did not cover the entire dune field. Consultation with the USGS in Flagstaff turned up a mosaic image of the Mariner data which covers the dune field completely and is oriented with north up (proctor.pgm, See Figure 5). The Viking images are also oriented with north up and having the same orientation greatly reduces the registration and preprocessing error. Having larger images of the dune field also gives QuakeFinder more context to lock on to. Using the new Proctor Dune Field mosaic, we ran two more experiments using the following parameters:

Experiment #3
QuakeFinder tile controls
tile-size 100 100
tile-skip 10 10
max-disp 20
precision 0
Experiment #4
QuakeFinder tile controls
tile-size 100 100
tile-skip 10 10
max-disp 50
precision 0

    These experiments increased the 'max-disp' parameter to allow the matches to be found across greater distances. This generated larger correlation surfaces to examine and possibly detect larger displacements.

Conclusions

    The mosaic images improved the coverage of both the dune field and the surrounding area; however, the results were very similar to those of experiments 1 and 2. This is most likely due to the great difference in resolution between these two images. This difference in resolution causes tile matches between the two images to be of lower correlation than would be expected of images of similar resolution. This low correlation can cause problems in the vector field such as those mentioned above. Given the results so far, we can conclude that there is no movement at or greater than the resolution at which we can detect it.

References

Crippen, R. (1992). Measurement of subresolution terrain displacements using SPOT panchromatic Imagry. Episodes,15(1), 56-61.

Stolorz, P., Blom, R., Crippen, R., Dean, C. (1997). Scalable scientific datamining on massively parallel computers. Space Microelectronics, 7, 40-49.

Stolorz, P., Cheeseman, P. (1998). Onboard science data analysis: Applying data mining to science-directed autonomy. IEEE Intelligent Systems,
September/October, 62-68.

Significance

The resolution and quality of the Viking and Mariner images available to us limits the scales at which we can detect ground motion. The currently ongoing Mars Global Surveyor mission will provide many more high resolution images of sand dunes on the Martian surface. MGS images taken at large time separations would allow us to detect the presence or absence of active sand dunes at the best resolutions currently available. However, it is possible that there are active dunes on Mars which change on longer time scales than are available during the MGS mission; therefore, high resolution MGS should form the basis of an image database to be used for on board motion mapping in future missions.

Points of Contact:

Paul Stolorz Bob Crippen
Jet Propulsion Laboratory Jet Propulsion Laboratory
Paul.E.Stolorz@jpl.nasa.gov Robert.E.Crippen@jpl.nasa.gov
(818) 393-5318 (818) 354-2475
Ron Blom Tim Stough
Jet Propulsion Laboratory Jet Propulsion Laboratory
Ronald.G.Blom@jpl.nasa.gov Timothy.M.Stough@jpl.nasa.gov
(818) 354-4681 (818) 393-5347

Expanded version of Science Directed Spacecraft Autonomy for Mars Missions