[HOT] Advice tracing footprints with poor imagery in Kathmandu, Nepal

Kate Chapman kate at maploser.com
Fri Feb 8 06:44:48 GMT 2013


Hi Ryan,

I'm with Maning I think you will actually have better success doing
the whole process manually. Also there is the issue of being able to
do the automated extraction with Bing. I think if you were to do the
automated extraction and then ask people to fill in you would be less
likely to get people to help. It can be frustrating to add buildings
when people have already added some of them, it doesn't allow one to
go systematically through each square.

What type of methodology are you looking for? I would suggest
assigning squares to people to do the tracing and then printing out
walking papers for field surveys to add the additionally attributes.
It might make sense to have two tasks. The first for everyone just to
trace the buildings and the 2nd for adding the attributes for the sake
of tracking. Actually having a couple different tasks in general might
be good. If you were to start in one area  you could correct your
methods based on what you learn from the first area to extend to the
2nd.

I would recommend having people that are better mappers on your team
look over the data very early on and correct mistakes. This will help
improve the later collected data.

Using the building plugin tool is really helpful. Also using features
in JOSM such as orthogonalize shape will improve your results.

One other thing to keep in mind is what analysis you are trying to do
at the end. Meaning I would think it is more important to get an
accurate count of the buildings and general size than make it exact.
You will still get a pretty good model of what would happen after an
earthquake.

Best,

-Kate

On Fri, Feb 8, 2013 at 1:09 PM, maning sambale
<emmanuel.sambale at gmail.com> wrote:
> There are opensource tools that can extract features from hires
> imagery, an example is GRASS, see this sample process (ignore the
> subjectline :))
>
> http://lists.osgeo.org/pipermail/grass-user/2007-August/040808.html
>
> But this requires, that you have direct access to the imagery (which
> isn't legally allowed with Bing)
>
> Another option for small areas is to try balloon/kite mapping:
>
> http://grassrootsmapping.org/
>
> In my experience, the most effective way is by manually tracing in
> JOSM (no computer algorithm can do what a human can see, at least not
> yet) and of course ground validation.
>
>
>
> On Fri, Feb 8, 2013 at 1:51 PM, Ryan Sommerville
> <ryan.sommerville at gmail.com> wrote:
>> Hi all,
>>      We're working on digitizing all building footprints in the Kathmandy
>> Valley. The current Bing imagery makes this a challenge in some areas,
>> especially densely packed neighborhoods.
>>
>> If possible, we'll utilize an automatic extraction of footprints and are
>> working on this now along with specific validation of that extraction. This
>> will still leave many buildings untraced (especially in dense neighborhoods)
>> so we'll also need to manually trace many buildings. The imagery sometimes
>> shows a clear building boundary but often it is unclear.
>>
>> Does anyone know of a methodology to accurately trace footprints using low
>> resolution imagery through a combination of mapping technique and
>> validation?
>>
>> Is it possible to add to the Buildings_Tools plugin to tag traces which will
>> require field verification?
>>
>> Any advice or suggestions is much appreciated!
>>
>> Thanks,
>> Ryan
>>
>> _______________________________________________
>> HOT mailing list
>> HOT at openstreetmap.org
>> http://lists.openstreetmap.org/listinfo/hot
>>
>
>
>
> --
> cheers,
> maning
> ------------------------------------------------------
> "Freedom is still the most radical idea of all" -N.Branden
> wiki: http://esambale.wikispaces.com/
> blog: http://epsg4253.wordpress.com/
> ------------------------------------------------------
>
> _______________________________________________
> HOT mailing list
> HOT at openstreetmap.org
> http://lists.openstreetmap.org/listinfo/hot



More information about the HOT mailing list