Just to throw more things into the mix, I have a third set of OS Street View tiles, only z11-z15 though as my overall method's much slower :-( which was produced using a different workflow to the first two. I'm using TimSC's warp-gbos program that was used to rectify the Scottish Popular Edition maps. This reprojects each source TIFF (and coverts it to JPG), it then tiles the JPGs, choosing individual pixels where going across source image boundaries.
I also see the characteristic kinks, but in different places. Mine seem to be generally *worse* (i.e. do not use mine for tracing!) although this might also be something to do with JPG compression. For some z16s I do have, compare: http://splintmap.geog.ucl.ac.uk/~ollie/osopendata/streetview/16/32760/21780. jpg and: http://a.os.openstreetmap.org/sv/16/32760/21780.png It looks like the os.osm set is correcting itself every few metres - hence those kinks appearing whereas I just have one big, horrible correction, about 70% down the tile. By the way you can see my set at: http://splintmap.geog.ucl.ac.uk/~ollie/osopendata/sv/ One thing I've already noticed, over and above the minor differences between the 900913 Street View sets, is how dramatically off the NPE maps are in certain places - GPS traces and OS Street View generally in agreement, while NPE can be off by ~2-300m. I redid a village last night which had been clearly traced in NPE - it was not good at all. None of the roads were connected to the other roads, and the GPS traces that were present for a couple of the streets were way off. I think the OS Street View-based tracing is now much closer to the truth. Ollie -------------------------------------------- Date: Wed, 7 Apr 2010 13:34:26 +0100 From: Grant Slater <[email protected]> Subject: Re: [Talk-GB] OS OpenData StreetView Tiles now available To: TimSC <[email protected]> Cc: [email protected] Message-ID: <[email protected]> Content-Type: text/plain; charset=ISO-8859-1 On 7 April 2010 12:53, TimSC <[email protected]> wrote: > In both the edgemaster's and Richard's tiles there are periodic kinks > (about 1 pixel in size) in what should be straight lines. It is more > obvious in Richard's tiles because they are less noisy. Any idea what is > causing that? > The warping between OSGB36 (EPSG: 27700) and EPSG: '900913'. Richard's tiles use a custom conversion script with antialiasing on the resample. Edgemaster's tile used gdal2tiles. I'm working on some cleaned up and pngcrush'ed tiles, but do not expect them in the next few days. The existing tiles are more than good enough. / Grant _______________________________________________ Talk-GB mailing list [email protected] http://lists.openstreetmap.org/listinfo/talk-gb
