2009/12/10 Peter Miller <[email protected]>: > > On 10 Dec 2009, at 14:18, Andy Allan wrote: > >> Hi Peter >> >> On Wed, Dec 9, 2009 at 9:20 PM, Peter Miller <[email protected] >> > wrote: >> >>> Thanks Andy, your wiki page is very useful and it is also good to >>> see the >>> imagery in Potlatch. Is the location of image encoded in the image >>> somehow, >>> or how else does PotLatch know where to load it? >> >> They are made into 256x256 tiles, with the same z/x/y.png notation as >> the main tileserver. It's the same way that the out-of-copyright maps >> work too. The !'s in the url on the wikipage are placeholders for >> potlatch to put in the tile numbers. For example, >> >> http://gravitystorm.dev.openstreetmap.org/imagery/stratford/14/8113/5397.png > > Thanks. That is very neat. > >> >>> When I tried warping the images I did feel that the algorithm used >>> by warper >>> was possibly a bit suspect - the warping suddenly went completely >>> wrong as >>> one added more control points in a rather counter-intuitive way. >> >> I'd discussed this with a few people over beers, and the problem >> appears to be that the warper's algorithm is assuming that you start >> with a fairly "flat" image, and doesn't take into account that the >> plane of the image might not be horizontal. It's for warping maps, >> after all, and we're semi-abusing it by warping photos instead. An >> extra control point or two can send it off wildly (as it tries to >> figure out how exactly a map sheet would end up stretched so) and of >> course it gets worse the further from vertical the photo is. >> > > So there is certainly a nice job for someone to sort out a suitable > algorithm that will work in practical situations when one doesn't > necessarily even have height data and certainly has distortions of > various sorts. The idea of associating 'ways' on the image to 'ways' > on the map sounds very neat assuming that one has a skeleton road > structure in place (which one could of course get from old OS maps if > necessary). I think it would then work a treat - certainly much better > than Warper currently manages.
I also agree would be great to be able to use linear features to rectifty, it's on my list! But I just thought I'd make you aware of the "advanced options" feature on Warper - by default, the algorithm changes based on the number of control points - you can override this behaviour. [You can also have a go at using the "Thin Plate Spline" method, which is best with lots of points - although ymmv and there may be a bug (mainly because no one uses it!). But essentially it gives a better warp when theres more points and treats the points as fixed - worth trying it offline with the command line GDAL tools if you have the time. ] **For all methods, the more points, the better.** Orthorectification is another thing though - thank god the photo's were in a relatively flat area! More generally as you know, in the world of aerial photography, there's usually extensive overlap (up to 60%?) for photos over areas and so you really shouldn't worry if 20% of the oblique image is unusable. We should try to get this large redundancy when flying again :) Cheers, Tim _______________________________________________ Talk-GB mailing list [email protected] http://lists.openstreetmap.org/listinfo/talk-gb
