Ok, after thinking about this a little and doing some more googling
about. It is hard to find the right terms for this question. Anyway this
is one approach that I thought of.
Given that you had a constrained area - what ever that might be.
You might generate horizon profiles of the area based on prominent peaks
or features, by viewing those areas from outside the area of interest.
The goal being to generate wireframe profiles of the ridges and
mountains in say N-degree steps about the prominent feature. These would
all get stored for future reference.
Now given a photograph, identify the horizon and any other significant
terrain features as wire frames.
Now try to match these against the sample references created above. This
will need to be done very approximately like just matching peaks and
allowing for horizontal spread based on point of view distance and
relative height differences from the reference perspective and the
camera perspective. The idea here is to narrow the field of possible
perspective images from 1000's to 100's.
This matching might be achieved by doing a linear regression that can
only distort your image profiles by stretching/compressing the image
horizontally and/or vertically and computing a least square fit against
the reference. You would want to keep the images centers aligned
right-left because you are trying to fine the reference image the best
aligns with your image because this will give you the heading onto which
you can then dos more detailed analysis. You can slide the images
up-down relative to one another. Or try to analyze matching prominent
features in the profile that match right-left from center.
Then a detailed analysis of this smaller number of possible viewing
angles and be analyzed. If you can uniquely identify 2-3 features, then
it should be possible to analyze the distance between them and their
relative heights and widths to compute heading, distance and azimuth of
the camera that you could then try to place more accurately on your DEM.
Anyway, having never done anything like this, this would be how I would
approach it without additional research to direct me in another direction.
-Steve
On 3/28/2011 5:27 PM, Michael P. Gerlek wrote:
Well, yes, I did do that first and have some angles on the more conventional
aspects of this, e.g. missile guidance. Being new to this area, though, I
thought I'd put out a query to see what else might turn up in the open
source realm (pure R&D being one thing; hackable code is something quite
different sometimes).
[that said, sometimes it's hard to even frame the right questions when one
is in a brand new area..]
-mpg
-----Original Message-----
From: Stephen Woodbridge [mailto:wood...@swoodbridge.com]
Sent: Monday, March 28, 2011 2:16 PM
To: m...@flaxen.com; OSGeo Discussions
Subject: Re: [OSGeo-Discuss] Finding position based on horizon profile?
On 3/28/2011 4:48 PM, Michael P. Gerlek wrote:
Consider the following hypothetical problem:
Assume we have a good elevation data set for a large region of the
earth -- say, an entire mountain range. Now let's say we have a
photograph taken from the ground, the horizon of which shows the
profile of a couple of the mountains in that range. Can you tell me
where the photograph was taken from?
Any pointers to research in this area would be appreciated.
Micheal,
Does this help?
http://www.google.com/#q=matching++"terrain+profile";
-Steve
_______________________________________________
Discuss mailing list
Discuss@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/discuss
