Topic: Orthorectification of high look angle (off nadir) images

Email correspondence with Sebastien below:

Don't bother with the coordinates. Once you have your two ortho images, in ENVI, use the Basic Tool -> Layer stacking, and stack your two images using cubic resampling and only the file overlap. It'll put back the images on the same grid, and then you can correlate the stacked images.
In the correlation tool, you may need to disable the "gridded" option.


On 10/16/2013 8:41 PM, William Armstrong wrote:
Hi Sebastien,

Thank you for your guidance last week. I have since gotten my hands on a
copy of LPS and produced DSMs from stereopairs and orthorectified each
image using its respective DSM and rational polynomial coefficient file.
The DSMs came out well (though TIN-y in areas of high relief or water)
and the orthoimages are coregistered to ~1.5 m (with relatively
consistent misregistration direction and magnitude over the image; our
signal is ~9 m in this time span, so I'm confident COSI-Corr will be
able to pick up the displacement).

However, the TIFFs I produce in LPS have a 0.5 pixel buffer around the
image, even though I tried to force the image corner coordinates to be a
multiple of image resolution. I am doing this in the 'Ortho resampling'
window in LPS, forcing the output cell size to a nice round number (say
1.00 m), and then 'align pixel'. The corner coordinates in that window
change to integers to reflect this change. However, when I output it as
a TIFF and bring into ENVI, the corner coordinates have a x.50 ending.
Obviously, this then creates a problem for obtaining a gridded output in
COSI-Corr. Have you experienced this problem? Is there any way to get
around it? I am currently running an ungridded correlation in COSI-Corr,
though I haven't seen one of these before, so I'm not sure how useful
the outputs will be.

Again, thank you very much for your help. I'm getting close to having
displacement vectors!


William Armstrong
PhD Student
Institute for Arctic and Alpine Research
Department of Geological Sciences
University of Colorado

web: instaar.colorado.edu/people/william-armstrong
email: william.armstrong@colorado.edu
mail: UCB 450, Boulder, CO 80309

On Oct 9, 2013, at 5:37 PM, Sebastien Leprince wrote:

Hi Billy,

LPS does a decent job with bundle adjustment, but you'll need
extensive manual editing to make sure no tie points are selected on
the moving parts.


On 10/9/2013 4:35 PM, William Armstrong wrote:

I look forward to seeing your talk at AGU. Thanks for sending along
the ppt.

I do have true stereo pairs, with the images acquired in the same
pass. Too bad the 3D code is too involved, but thank you for the
information. I will move on pre-processing the imagery with other
software as you suggest. I imagine, though, that I will not be able
to coregister the images as well using the Intergraph software? I
have biweekly imagery, so the glacier may have only moved ~5 m
between images, so it seems having rigorously coregistered imagery is
necessary. Alternatively, I can just take larger temporal steps.

Thank you for your guidance,


On Oct 9, 2013, at 1:16 PM, Sebastien Leprince wrote:

Hi Billy,

The stripping comes from a bug in COSI-Corr because the public
version is really old. We should post a new version soon.

Regarding the sensitivity to vertical offset and parallax, you will
notice its effect as soon as the angle is larger than 3-5 degrees.
Thus, for your processing, you need to do a full 3D formulation.

For the full 3D formulation on glaciers, you need "real" stereo
acquisitions, meaning stereo images acquired simultaneously from the
same track, not multi-temporal images. Is it what you have?

Even if I wanted to, I wouldn't be able to share the code for 3D
extraction. The complete code is different, it takes care of
complete bundle adjustment of all imagery simultaneously, and then
the correlation has to be different including multiscale and
regularization. IDL can't really handle that kind of processing and
our new codes are optimized in C++ for our server. Codes aren't
currently portable and full processing takes a few days.

An alternative solution you can do for now is to use ERDAS LPS,
produce DSM and ortho-images for each date you have the
stereo-pairs. Then use COSI-Corr to correlate the ortho-images.

We're working on putting together a web/cloud interface to our 3D
code, but it'll take a long time.

Here's a presentation on Glaciers using Worldview, you can see first
results in New Zealand glaciers on slides 17-24. I'll give a more
detailed presentation at AGU in December about the principles:

ftp://ftp.gps.caltech.edu/pub/leprincs/ … _2013.pptx


On 10/8/2013 6:16 PM, William Armstrong wrote:
Hi Sebastien,

I would have posted this on the forum, but I have included some
imagery that I don't think I could have posted on there.

I saw your comment on the COSI-Corr forum that correlating imagery
with a high incidence angle is a full 3D problem and that the code
is not publicly available. I am trying to obtain velocity fields
from correlation of WorldView imagery for the Kennicott Glacier in
southeastern Alaska. I have been able to obtain a good amount of
WorldView imagery, but all the scenes are shot at relatively high
off-nadir view angles (minimum mean off nadir view angle = 14 deg;
max mean off nadir view angle = 28 deg). Are these large enough off
nadir that we have to solve the full 3D problem? I have stereo
pairs for all of the images shot this summer. Our WV imagery is
~0.53 m pixel ground resolution and we have images shot every two
weeks, which corresponds to ~2 to 5 m displacement. Would you be
willing to share the code to solve the full 3D problem if this is
an issue? I am a PhD student at University of Colorado working with
Bob Anderson.

While I am writing you, I will take this time to ask a processing
question as well. Some of my images produce strange striping when I
go to orthorectify and resample them in COSI-Corr. I have attached
two images so you can see what I mean. One is the full scene and
one is a zoom in. I retried the orthorectification without a DEM
and no striping occured, so it seems the DEM is the issue, but I
have orthorectified images with the same DEM without this strange
striping pattern.

Here is the image information: Sensor = WV02; Original pixel =
~0.53 m; Resampled pixel = 1 m; Off nadir = 23 deg
The DEM is ASTER derived, with 15 x 30 m pixels. It covers a much
larger extent than the scene shown.

Do you know what might be causing this striping?

Thank you very very much.


Billy Armstrong

William Armstrong
PhD Student
Institute for Arctic and Alpine Research
Department of Geological Sciences
University of Colorado

web: instaar.colorado.edu/people/william-armstrong