[OSM-talk] Google Maps comparison
dair at refnum.com
Thu Oct 11 19:50:56 BST 2007
David Earl wrote:
>>Not to knock Dair's work, which is stunning, but I'd be wary about
>>announcing too many errors without some specific proof, lest it be
>>seen as crying wolf.
>Indeed. I have found numerous OSM errors while connecting to
>existing mapping in my area, and made many myself
Likewise - I want to be clear that my intent with this page
isn't to say "look how terrible Google Maps/TeleAtlas data is,
OSM is perfect".
OSM data contains errors too; Haywards Heath was unmapped when I
started, and I have been amazed how often I missed things when I
went back and re-surveyed an area.
So although it's a difficult question how you measure
accuracy/completeness, I think a comparison like this is a
useful way to start.
As far as proof goes, I think this is really the only way to do
it (build an OSM map, look for differences, then re-survey as needed).
>Of course the biggest OSM error is that lots of streets are missing
>altogether. Often this is obvious because there's just a skeleton,
>but sometimes people have mapped 2/3 of streets spread evenly in a
>village (i.e. not the west half of the village, say, but an
>apparently random distribution which makes it look plausible).
It would be good to be able to produce a kind of confidence map
for OSM, indicating how sure we were that an area was "done"
(whatever that means).
Percentages imply too much precision, but a 3 or 4 point scale
where you can say "nothing", "draft", "good", "complete" might work.
Complete doesn't mean error-free, it just means that this map is
at least as good as any other map you're going to find for the area.
>So long as we can't say in some way that we believe this is a
>complete area, I don't think it makes sense to talk about errors.
That's true to an extent, but I think you can start talking
about errors or completeness as soon as you identify a specific
area to talk about.
Ultimately countries are made out of counties, which are made
out of cities, which are made out of districts, which are made
out of streets (or similar).
If you go far enough in you can always find some area you can
say is "done", and although completeness drops as you zoom out
there is a range between individual streets vs the entire planet
where you can say "this bit is done, this bit needs more work".
It would be an interesting experiment to let people place shapes
on an OSM map to indicate how complete they felt an area was.
-dair (by "complete", I just mean "captures some topology" -
house numbers, turn restrictions, etc, means the process never
dair at refnum.com http://www.deathvalleycycle.com/
More information about the talk