[OSM-talk] People's Map

Tim Waters (chippy) chippy2005 at gmail.com
Sat Apr 11 00:47:46 BST 2009


2009/4/10 andrew heggie <list at sylva.icuklive.co.uk>:
> On Friday 10 April 2009 22:05:15 Martin Spott wrote:
>> D Tucny wrote:
>> > How much does a small plane with camera mount cost to hire for a day? :)
>
> I don't know about a day but 17 overlapping images of an approximately 10 by
> 10 km area cost me GBP600. Once rectified they were not true "plan" images
> being slightly oblique. I guess the best images would be taken higher ( mine
> were 14000ft IIRC) with a longer lens to reduce parallax errors.
>>
>> It depends on wether you're going to invite the pilot for a nice
>> meal  :-)
>> I think the most tricky part of the story is still to rectify and
>> adjust the resulting aerial imagery. If you managed to develop a
>> procedure for this step, please let me know  ;-)
>
> Jukka Rahkonen showed me how to do it with gdal_translate and gdalwarp. It's a
> lot easier than it looks at first.
>
> Essentially the first bit burns ground control points into the image the
> second then stretches the image and produces a geoTIFF from it. I visited 4
> ground control points with my GPS each about 3km apart and at prominent
> points near the corners of each image. I suggest you need farm ore ground
> control points than this at this scale because my georectified image was up
> to 20 metres adrift in some parts.

Indeed it's quite easy, at the basic level you can use a similar
service such as http://warper.geothings.net or choose from a desktop
GIS, most of them have some way of doing this. These would georectify
images, but we should orthorectify them too which is a bit more
trickier.

What happens is that the distance from lens to ground is different
over varying terrain, so it doesn't match what a map would be.

Crudely, imagine taking a snapshot photo from the plane just as you
fly over a mountain top: way down below you'd see the tiny roads, but
most of the frame would be the mountain peak. Of course,
orthorectification is more important with terrain at different
heights, and less so for flat ones.

I think the way these are done are to use a digital evevation model -
I would hope the free SRTM could be sufficient?




More information about the talk mailing list