[Talk-GB] Stitching Aerial Photographs

Matt Williams lists at milliams.com
Sun Sep 20 23:08:13 BST 2009


2009/9/20 John Robert Peterson <jrp.crs at gmail.com>:
> now hugin/panarama tools/sift has a strong concept of control points, so
> does warper -- is there any way that we could get them to use the same
> points, and bolt the 2 together automatically?
>
> nearly a week ago I posted the following, any further thoughts on it:

I must have missed that email.

> The ideal situation for me would be if we could have some automated tool on
> a sever somewhere receiving images from users, and automatically rectifying
> them.
>
> This sounds imposable, but I believe that with a very small amount of user
> help it could work:
> Images can be automatically pinned together in overlaps (and high res images
> can be pinned onto wide area images) using tools similar to those in
> Panorama Tools / hugin (this searches for notable points in the image data,
> and matches them between images);
> searching the images for road markings (and/or cars) and matching these to
> the gps traces already uploaded would give a reliable enough way to add real
> world control points;
> if a concept of altitude is added, automatic control points between images
> appearing on top of skyscrapers would not be too damaging;
> approximate terrain relief for most of the planet is already known, this can
> be tied into the above to help;
> manually adding control points would be done by displaying already rectified
> images next to a map, with the user adding pins to useful points;
> images with no rectification data at all would be added using a push pin
> system similar to that used on http://warper.geothings.net/
>
> the above is a pipe dream at the moment, but has the potential to
> revolutionise how we do things in my opinion.

This sounds like it could be very useful if it turns out to be
possible. However, rather than Hugin/ Panorama Tools I think that
something like Bundler would be much more resiliant and automatic.
OpenStreetPhoto mention that they're planning on using Bundler but I
don't believe that they've actually done anything with it yet.
However, it would (theoretically) allow all the images to be put
together in a 3D space and create a sort of textured 3D model of the
area. This could then be projected down onto a DEM or a flat plane for
tracing over. It would essentially solve the ortho-rectification
problem and only require us to align the image to coordinates.

So, there's another pipe dream for you :)

-- 
Matt Williams
http://milliams.com




More information about the Talk-GB mailing list