[OSGeo-Discuss] Anyone openly georeferencing maps and pictures in time?

Tim Lund tim at lund.co.uk
Sun Aug 3 05:00:09 PDT 2014

First I'd like to say a big thank you to Pat, Eli & Bob for your responses
here, and also to Jeff who I originally emailed, and suggested I email this
list.  Problems of this type clearly are technically soluble, and some, e.g
in the military context, which I'd not been thinking about, have been
solved.  So let's try to redefine what I looking for, and the questions
which arise.  So:

1. A process for geotagging images which are sparse in both time and space,
but with human expertise on hand.

I think this differentiates the problem from the Structure from Motion
problems linked to, although I note that they include looking at sparse
data.  It's just that having just the occasional photos and drawings
surviving of a location at various points in the past take sparseness to
another level.  I may be wrong, but there has to be a sparsity threshold
beyond which you need to design in the human expert.

For this stage, I think now I was confusing myself by thinking about
warping images to maps; more sensible to warp maps to images, so having
tentatively tagged an image with a location, direction of view and
magnification, then call up a warped version of vector layer map, such as
OpenStreetMap, to compare the edges on the image with what you'd expect to
see, and adjust the initial estimates for the image geotagging.  It
wouldn't need to be perfect, just recognisably making sense to the human
researcher; it's not as if we want to land a missile on anything with
pinpoint accuracy.

An incidental thought here - what does OpenStreetMap do when new roads and
buildings come along?  Can it be rolled back to how things were?  If
there's a data structure of that, could that not be extended to how streets
looked before anyone thought of OpenStreetMap?  If I was trying to locate a
historical photo, it could be helpful.

2. A data structure for holding such geotagged images and relating them to
other types of entity.

Thanks to looking videos showing architectural drawing packages (SketchUp)
interfacing with Google Earth, in which buildings are considered as an
entity type 'component', I think that's the sort of thing I want.  I'm
thinking of this in the context of how city-scapes change, so that seems
natural.  Obviously if it was landscapes, components could be water
features or other components of the non-built environment.  So you'd want a
relation to say that an image is of some component - which would have its
own geotag, and also some vector data to describe its 3D extension based on
the edges obervable - by machine with expert guidance.  This would be a
many to many relationship.

3. A process to use this data to realise what it would look like if you
looked a certain way at a given point in time - so this is where you would
warp the images rather than the maps.  This also has to be a solved problem
- architects and town planners must be doing it every day as they visualise
projected developments, and are also able to add texture from existing
images.  The product I was just looking at in this connection is SketchUp

4. Making it happen.

I'm not going to write too much about this, because it would be too vague
and conjectural, but any thoughts on what it would take to achieve the
above, and additional ideas on the sort of entities who might argue for it
and fund it would be appreciated!  I'm also thinking about copyright issues
... any thoughts on that welcome too.


On 1 August 2014 20:01, Pat Tressel <ptressel at myuw.net> wrote:

> Tim --
> This is a project idea which seems obvious to me, and one which would so
>> obviously benefit from OSGeo involvement, that I feel someone on this list
>> will know very quickly if anyone is working on it in an open data way.  It
>> comes from thinking about the warping which needs to be done to get from an
>> aerial photograph to a map, and extending the thought to what can be done
>> with a very oblique image - such as I might take standing on the ground.
>>  Any photo, not just an aerial one, can be considered as a map just waiting
>> to be tagged with scale,  projection, geolocation and date.  The photo
>> doesn't have to be great quality - perfection is not needed.  In fact, if
>> we allow some artistic licence, we could apply the same process to scans of
>> historic prints and paintings.
>> And if we had a library of such geotagged images, researchers would be
>> able to specify an area and a time range, and search for images whose area
>> of coverage overlapped it taken during the given period.  It would be of
>> antiquarian interest - there's an organisation I belong to called the London
>> Topographical Society  <http://www.topsoc.org/front/index> which has
>> access to a mind-boggling number of maps, old photos and prints of London -
>> but also to academics in Geography and Town Planning departments.  It would
>> also be of commercial interest to developers looking at the planning
>> context for new developments.  And I think I've read somewhere of
>> commercial companies - Google, Facebook? - collecting various picture of
>> the same location, e.g. a holiday destination, and using the combined data
>> to produce images with unwanted obstructions eliminated.  It has to be
>> possible, so is anyone working on developing an open source library of
>> images so tagged?
>> Brief background on me; I'm a maths graduate, now approaching retirement,
>> and with interests not only in history, but also urban development, so a
>> project along these lines is something I'd love to get involved with.
>>  Although I might dream to doing some coding, that's just not realistic
>> when my skills are more in MS Office applications and VBA.  I've also been
>> looking at 'R' and QGIS, and I could get to the point of doing the tagging,
>> except for date stamping, but if there was anyone else further up the
>> learning curves for these, it would be good to link up.  I also have a lot
>> of possible contacts with people who might be interested in such a project
>> as users, which would also make a difference.
>> It seems like such a nice project, so hoping someone can help
> This is a popular area, since it relates to side-scan sonar, side-looking
> aerial radar, and cameras suspended from drones, which, even if they're
> intended to be pointing down, rarely are.  (It also seems to be somewhat a
> solved problem, just not open-source -- as you may guess, this has military
> use.)
> It also comes up in autonomous vehicles, since one wants to infer (for
> instance) distance of objects from imagery.  For both this and
> georeferencing, sequences of (partly) overlapping images -- video -- are
> very useful.
> In fact, this subject is under current discussion over on the Humanitarian
> OpenStreetMap Team mailing list, with the revival of OpenAerialMap.  I'd
> recommend joining up with the folks over there.
> I'm CCing some folks (Stephen Mather, Kate Chapman, Michael Patrick) who
> are involved with OpenAerialMap and / or OpenStreetMap and georeferencing
> in general.
> Some references re. georeferencing imagery (specifically for drones) and
> related (Michael probably has more):
> http://opendronemap.github.io/odm/
> http://dronemapper.com/  (commercial image processing service)
> http://flightriot.com/
> http://ccwu.me/vsfm/
> -- Pat

151 Silverdale
SE26 4SQ

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/discuss/attachments/20140803/0d33cf46/attachment.html>

More information about the Discuss mailing list