27.08.2012 20:12, Frank Warmerdam пишет:
On Mon, Aug 27, 2012 at 11:50 AM, Even Rouault
<[email protected]> wrote:
Making algorithm work with imagery that is piece-wise loaded can be very
complicated indeed. And if you try to correlate images that are taken by sensors
that have very different angles, then you could need to correlate parts that are
not at all at the same position in the 2 images.
Even,
I think it will actually not be too hard to remove this problem
with loading the whole image. It seems that the reference
point objects (the somewhat poorly named GDALFeaturePoint)
stores a "descriptor" representation of the point with it and so
after these are collected from one of the images it is not
necessary to keep the original image in memory.
I'm still not sure how to handle the image scaling and
conversion to a luminosity image more smoothly.
Best regards,
There are several ways to use not the whole images (prevent load full
image to the memory).
1) Use aerial images metadata (e.g.
ftp://www.aerialrobotics.eu/winter-high-overlap/trace.txt.TRIG1-001.pix4d.dat)
to get overlapped areas or part of this areas
2) Use overviews to gather points, and make their placement more
accurate on full size image parts
3) May be OpenCL or some parallel threads
4) Gather small amount of points, and use them to make inaccurate
orientation, and after make more accurate feature points gathering on
overlapped areas, etc.
But I think, that by now the block images (>2) points adjustment is next
step to create single seamless image.
Best regards,
Dmitry
_______________________________________________
gdal-dev mailing list
[email protected]
http://lists.osgeo.org/mailman/listinfo/gdal-dev
