[gdal-dev] Guidance sought on using gdal-translate

[Richard Greenwood]

On Fri, Aug 1, 2008 at 8:02 AM, Frank Warmerdam <warmerdam at pobox.com> wrote:
> Christopher Hunt wrote:
>>
>> Hi Frank et. al,
>>
>> Sorry to be a pest. I've been trying tons of things, trawling tons of
>> search results, and I'm really no better off than I was a few days ago. :-P
>>
>> In summary I have an aerial photo image of an airport with three well
>> known points (expressed in lat/longs). I'm wanting the image to be projected
>> using a equidistant cylindrical projection.
>>
>> On gdal_translate's gcps, I'm presuming that the origin of the image is
>> assumed to be top-left for the gcp pixel and line values... and that the gcp
>> easting and northing values are expressed in metres?
>
> Chris,
>
> Correct.
>
>> How do I centre a projection? lat_0 and lon_0 values?
>>
>> Thanks again for your time - I appreciate that time is precious.
>
> I honestly believe that reprojection is the wrong tool to take a raw
> aerial photo and try to turn into an orthophoto.  Normally this process
> is address by orthorectification software that takes into account the
> geometry of the camera lens, the orientation of the camera when the photo
> is taken, and the elevation model of the target region.
>
> If you are going to try and accomplish orthorectification by something as
> simple as reprojection then the fit is going to be relatively poor in all
> but the most ideal circumstances.
>
> So, sorry to say it, but I'm not going to wrack my brain to try and solve
> a problem that I think is likely insolvable.

I'm coming into this thread pretty late, but I think Frank is on the
right track. You can't assume that your photo is orthorectified. If it
were, you probably would not need to provide GCPs because control
points would have been required to rectify it in the first place.

Does the image have a collar, possibly with fiducial marks (usually 8
small "+" marks at the edges) and other info? If so it was taken with
a "real" photogrammetric camera. If the collar edges are wavy (not
perfectly straight and square), then the image may indeed have been
ortorectified and you probably can simply project it. If the collar
edges are perfectly straight, then it probably hasn't been rectified.
GRASS has a great ortho rectification module for working on single
images.

If your image does not have a collar, then you are really guessing.
And in any case, if you can get some info from who ever produced the
image, you could greatly reduce the guessing.

Rich

-- 
Richard Greenwood
richard.greenwood at gmail.com
www.greenwoodmap.com