8.19. Clementine

This example shows how to build CSM camera models for the Clementine Near Infrared (NIR) camera, bundle-adjust an along-track sequence of frames, and mapproject the sequence onto a reference lunar terrain model.

Clementine flew four framing cameras (UVVIS, NIR, HiRes, and LWIR) in a near-nadir lunar mapping orbit in 1994. The mission was designed for global color mapping, not for stereo, so the along-track convergence between frames is small [CSRW02]. Coarse Clementine stereo topography has nonetheless been produced from the parallax between adjacent frames, at about 1 km per pixel and a few hundred meters of vertical accuracy [CSRW02].

This example does not attempt creating a stereo terrain model. It demonstrates that Ames Stereo Pipeline can process Clementine images and cameras, and it shows how well the uncontrolled Clementine pointing georeferences the imagery onto a more recent lunar DEM.

The NIR frames used here are 256 by 256 pixels, with a ground sample distance of about 230 m/pixel.

8.19.1. Optical distortion

The ISIS Clementine NIR camera applies a single-parameter radial optical distortion. As of ALE version 1.2.0, the Clementine driver assumes zero lens distortion. The effect is minor, on the order of a fraction of a pixel, and this will be fixed in a future ALE release. The CSM cameras in this example were created with development ALE code that supports the Clementine NIR distortion, so the CSM camera matches the ISIS camera model at subpixel level.

8.19.2. Fetching the data

This example uses a sequence of consecutive NIR frames from revolution 284, over the lunar nearside near longitude 44 degrees East and latitudes 10 to 19 degrees South. The frames overlap along-track by about half a frame. They can be found with the Orbital Data Explorer, searching for the Clementine NIR EDR product type.

Set the base url:

base=https://pds-geosciences.wustl.edu/geocopy/imaging/clem1-ley-abuhln-2-edr-v1.0/cl_0066/lun284/lnxxxxxx/lnxxxxxh

The xxxxxx strings in the path above are literal directory names in the PDS archive, not placeholders to fill in.

Download a handful of images:

for id in lna1571h lna1604h lna1637h lna1670h lna1703h \
          lna1736h lna1769h lna1802h lna1835h lna1868h; do
  wget $base/$id.284
done

8.19.3. Preparing the data

Ingest each raw frame into an ISIS cube, apply the radiometric calibration, remove detector noise, and attach the SPICE data. The calibration and cleanup use clemnircal, clemnirnoise, and clemnirclean. This needs the clementine1 ISIS data area, including its calibration files, and the mission kernels, fetched with downloadIsisData under $ISISDATA (Section 2.3.1):

for f in lna*.284; do
  b=${f%.284}
  clem2isis    from=$f             to=${b}.raw.cub
  clemnircal   from=${b}.raw.cub.  to=${b}.cal.cub
  clemnirnoise from=${b}.cal.cub   to=${b}.noise.cub
  clemnirclean from=${b}.noise.cub to=${b}.cub
  spiceinit    from=${b}.cub
done

The intermediate files .raw.cub, .cal.cub, and .noise.cub can be deleted after this step.

Fig. 8.66 Twelve Clementine NIR frames, after calibration and noise removal, contrast-stretched for display, in along-track order (left to right, top to bottom). Adjacent frames share about half their area.

8.19.4. Creating the CSM cameras

Create a CSM camera (an ISD .json file, Section 8.12) for each cube. The -k option furnishes the kernels recorded by spiceinit:

for c in lna*.cub; do
  isd_generate -k $c $c
done

The cub is passed as $c twice, with the first being a value for -k.

8.19.5. Bundle adjustment

Refine the cameras with bundle_adjust. Because the frames were acquired for color mapping, the baseline between them is small and the rays are close to parallel, so a very small minimum triangulation angle is used, and a forced triangulation distance (in meters, about the spacecraft slant distance) places the triangulated points when the angle is below that threshold. The camera position uncertainty is set large, as the absolute accuracy of the Clementine pointing is assumed to be rather uncertain. A large number of interest points are requested:

ls lna*.cub > images.txt
cat images.txt | sed 's/\.cub$/.json/' > cameras.txt

bundle_adjust                             \
  --image-list images.txt                 \
  --camera-list cameras.txt               \
  -t csm                                  \
  --ip-detect-method 1                    \
  --ip-per-tile 5000                      \
  --matches-per-tile 2000                 \
  --min-triangulation-angle 1e-10         \
  --forced-triangulation-distance 617841  \
  --camera-position-uncertainty 1000,1000 \
  --robust-threshold 0.5                  \
  -o ba/run

Deriving the camera list from the image list with sed keeps the two in the same order, as required.

This is a purely relative adjustment. It makes the cameras consistent with each other, but does not tie them to the ground. The median reprojection error after adjustment is about 0.16 pixel, from about 80 matches between adjacent frames. The convergence angles between the frames are small, about 3.5 degrees between adjacent frames, rising to a few tens of degrees for frames further apart in the sequence (Section 16.5.11.4).

For CSM cameras, bundle adjustment writes an adjusted camera state file for each camera, ba/run-<camera>.adjusted_state.json. This is the CSM camera with the adjustment applied inline (Section 16.5.11.7). These files are passed directly to the tools below.

Fig. 8.67 Interest point matches (red dots) between two overlapping NIR frames, lna1637h and lna1670h. The shared craters are matched well.

8.19.6. Mapprojection onto a reference DEM

To see how well the adjusted cameras georeference the imagery, mapproject the frames onto a lunar reference DEM and overlay them on the DEM hillshade. Here the reference is the global LOLA DEM at 128 pixels per degree that ships with ISIS, under $ISISDATA/base/dems. The SLDEM2015 LOLA and SELENE terrain camera merge (Section 8.14) is a higher-resolution alternative for latitudes within 60 degrees.

Set the projection string of the reference DEM:

proj="+proj=eqc +lat_ts=0 +lat_0=0 +lon_0=180 +x_0=0 +y_0=0 +R=1737400 +units=m +no_defs"

Crop the reference DEM to the region and make a hillshade:

dem=$ISISDATA/base/dems/LRO_LOLA_LDEM_global_128ppd_20100915.cub
gdal_translate                               \
  -projwin -4310714 -242823 -3885643 -667765 \
  $dem lola_crop.tif
gdaldem hillshade lola_crop.tif lola_hs.tif

Mapproject each frame with its adjusted CSM camera, at the native ground sample distance, into the DEM projection, then mosaic the results:

mkdir -p map
for cub in $(cat images.txt); do
  prefix=${cub%.cub}
  cam=ba/run-${prefix}.adjusted_state.json
  mapproject            \
    --tr 237            \
    --t_srs "$proj"     \
    lola_crop.tif       \
    $cub                \
    $cam                \
    map/${prefix}_map.tif
done
dem_mosaic map/*_map.tif -o map/strip.tif

Fig. 8.68 The mapprojected Clementine NIR strip drawn over the LOLA hillshade of the same area. The craters in the strip line up with the same craters in the hillshade on either side, so the adjusted cameras register the imagery to the reference terrain well even without any additional alignment.