ossim news

Registration Plugin

Thu, 4 Sep 2008 08:49:23 -0400

Notes from Hernán De Angelis

I want to make clear first that all I have done is in a very
early experimental stage, based on what I understood of the
registration manuals, my own experience and your help in this list.
Therefore, what I write below should *not* be considered as the
canonical rules of how to work with OSSIM registration.

Together with two colleagues I drive a glaciological project in
Patagonia (Southern South America, thus the need for south UTM
support). We have a bunch of satellite images (Landsat MSS, TM and
ASTER) and some old scanned aerial photographs. The purpose of using
automated registration is to overcome the difficulties/drawbacks of
manually picking up ground control points and to get the best possible
registration.

One of the major difficulties of using automatic registration over
glaciated areas is the fact that glaciers flow and some features over
them are displaced over time. In fact, this principle has been used by
glaciologists for a long time to track features and estimate surface
velocities in glaciers (see IMCORR:
http://www.nsidc.org/data/velmap/imcorr.html). However, glaciers in
our area are relatively small and occupy only a fraction of the image,
so we hope they are not introducing severe errors (though this might
not locally be true). In fact, it would be nice to have the
possibility to mask glacier areas, so they are not included in the
correlation (feature request? mmm may be).

We use one reference ASTER image (we think the original georeference
is excellent), with 15 m pixel size. This image is actually a subset
of a mosaic of two, because our interest area was not covered by only
one image. The mosaic was done with ERDAS. All other satellite images
were cropped to cover the same area as the reference image and also
resampled to 15 m. The aerial photographs were individually geocoded
and orthorectified in GRASS with an SRTM model. The elevation database
is composed of SRTM tiles and covers completely the area of interest.

I start with "correl". The slave images are always geocoded and more
or less within 10-30 pixels of mismatch. Normally I use it in this
way:

correl -m master_band -s slave_band -c 0.90 -d 0.001 master.tif slave.tif

during a registration I might vary the values of -c and -d, but those
two (0.90 and 0.001) are giving good results with most images. I
always try to choose master and slave bands that are radiometrically
similar. Examples: ASTER VNIR 3 with LANDSAT TM 4 or MSS 3. Actually,
I do not know how sensitive the algorithm is to radiometrical
differences in images, but as far as my experience is concerned the
visible near infra-red bands may be interchanged without much trouble
(this is hardly surprising). An interesting issue is that registering
a resampled MSS image to the ASTER reference presented no serious
problems.

After this I work mostly with "modopt". I simply do:

modopt tiepts.xml

(that is with the default polynomial model) and get generally very
acceptable results. Only eventually I might need to do rejout:

rejout -a 2 -i 0.5 tiepts.xml

I follow the manual after this: I change the name of the resulting
.geom file to add the base name of the image (e.g. modopt.geom ->
image.geom). I open both the reference and the slave images in
imagelinker and do a visual control. If I am happy I switch to full
resolution and save the results as a new image. WARNING: before doing
this I must normally tell imagelinker that the UTM is south (somehow
OSSIM is not happy with the Southern Hemisphere). I do this in Edit ->
View.

According to the manual the process should now be finished. However, I
noticed that images produced in this way cannot be displayed correctly
by, for example, OpenEV. I therefore do:

orthoigen --utm image1.tif image2.tif (where image1.tif would be the
image saved by imagelinker)

to get the metadata correctly written in the geotiff. However, even
after this step, gdalinfo does not report South Hemisphere.

The procedure with the aerial photos is different. The aerial photos
are reported to have abot 4 m ground resolution. After some (not very
lucky) experiments with reduced resolution datasets I used the full
images, which cover only a tiny portion of the reference image:

correl -m 0 -s 0 -c 0.85 -d 0.002 master.tif aerial.tif

Notice that now the correlation coefficient is smaller. That was the
only thing I had to do to make it work more or less acceptably. Modopt
was used to optimize the model (polynomial). Errors in the aerial
photos include severe local distortion near the edges (associated with
traditional radial errors that could not be eliminated during the
orthorectification) and some displacements of relative high peaks.
Otherwise it was reasonably OK.

My results until now can only be said to be very acceptable. They are
not perfect, but I am working on it!

Regards,

Hernán

Palomino3D

Fri, 15 Feb 2008 08:08:26 -0500

Jim Brooks has developed an excellent open source flight simulator based on OpenSceneGraph. Palomino3D runs on Linux, FreeBSD, Mac and Windows. Jim lives in Melbourne Florida and discovered that we were in the area as well. After meeting for lunch Jim added ossimPlanet into his flight simulator. Palomino 3D is written in C++ and uses the CMake build system to support multiple platforms. We’ve added some of the screenshots to the gallery.

Palomino3D Screenshots

ossimPlanet 1.7.0

Sat, 17 Nov 2007 17:55:33 -0500

Binary installers are now available for ossimPlanet 1.7.0. This is the first release where we are supporting Windows XP installers in addition to the MacOSX versions. Some of the added features in this version are:

3D Compass - an OSGeo compass for heading and tilt
KML and KMZ report including 3D Collada Models
Dramatically reduced CPU usage
BigTiff support
Full compliment of GDAL formats supported
Improved WMS interfaces and dialogs
Synchronization of Navigation, Data URLs, and OGC WMS sites between distributed planets
New Router/Action mechanism for upcoming server support
Support for Precision Lightworks 3D Urban Models
Numerous maintenance fixes
Plugin support for the Pixia NUI format
Updated documentation
Bundled installation for Windows and Mac
Bundled global elevation, bathymetry, and blue marble NG backdrops in full install, executables only in minimal install

Join the ossim mailing list and/or the #ossimplanet IRC channel at irc.freenode.net for more information. Additional information at http://www.ossim.org

Downloads at http://ossim.telascience.org/ossimdata in the Windows and Mac subdirectories.

BigTiff support in OSSIM

Fri, 12 Oct 2007 10:58:00 -0400

BigTiff support is now available in the OSSIM baseline. BigTiff is an extension to the tiff libraries that eliminates the 2 or 4 gig file system limit associated with tiff files. Libtiff 4.0 now includes this support and the library has been integrated in to OSSIM. Extensive testing has been performed on files up to 100 GB with OSSIM and ossimPlanet. OSSIM can now build BigTiff files making it possible to mosaic large areas and output the product as a single large file. Using the tiled multi-resolution structure supported by bigtiff, ossimPlanet can fly through terrabytes of linked imagery.

Collada Models added

Fri, 12 Oct 2007 10:53:18 -0400

Improved kml and kmz support is now in ossimPlanet. This includes the addition of Collada models from the Google Earth 3D warehouse. Necessary software is now in the svn repository, planning on new binary packaging at the end of the month. These models are supported through the Collada plugin in OpenSceneGraph 2.2.