Okay, well for a quick and dirty, but hopefully fairly randomized
approach to randomly sampling and classifying the data against Bing
imagery, I wrote a short interactive bit o' Python:
https://gist.github.com/2655895
If you're bothered about the data for mini-roundabouts, could you give
it a spin please? The code might be useful for other tag wars too,
prod me if you want it expanding to ways and closed ways.
For a random sampling of 100 nodes tagged highway=mini_roundabout from
Josh's set, I visually classified as
1 "tc? island traffic calming of some sort"
2 "parking" (either an aisle
10 "j" (ordinary junction: note that imagery could pre-date a
roundabout being built)
15 "?" (can't tell: Bing imagery is too blurry or it's not clear)
19 "tc" (turning circle at the end of a road, with or without a
solid centre)
23 "r" (solid-centred roundabout you physically cannot drive over)
30 "m" (paint- or bump-centred mini roundabout, "traversable")
This is the sample-out.put.csv from the gist. That's a worryingly flat
distribution: the data is fairly mixed, and there's nearly as many
turning circles (which *really do* need tagging as
highway=turning_circle) as the (IMO) wrongly classified
non-traversable "miniature roundabouts". And a worrying number of
plain T- or cross-junctions. The predominant usage is for the
paint-centred type, but they seem to be somewhat limited by country.
It also becomes fairly apparent that personal biases can easily come
into play when visually evaluating how a tag is used over *the entire
world*, which is why more eyes are probably a good idea. FWIW, I
tended to "see" more flat or bump ones ("m" in my scheme) in GB,
France and Germany, perhaps because more of the corner cases fit my GB
expectations better. But I'm hoping my observations will be borne out
by others', and that it's about as un-subjective as you can get with
this sort of thing.
--
Andrew Chadwick
_______________________________________________
Tagging mailing list
Tagging@openstreetmap.org
http://lists.openstreetmap.org/listinfo/tagging
