[OSM-talk] Suggestion for an Unconference
ben.last at nearmap.com
Mon Nov 29 00:50:44 GMT 2010
On 27 November 2010 06:25, Ed Avis <eda at waniasset.com> wrote:
> For addressing, I guess it is usually sufficient to have a street name -
> exact addr:housenumber stuff is not needed I assume?
Something I can comment on (speaking as just me, not nearmap for once!)
having been looking recently at some feedback from people relating to
searching done using OSM data; unfortunately street isn't good enough. As
SteveC points out in a later reply, some places have roads that are very
long; the Albany highway in WA is 410km long and the numbering runs from
around 51 in Albany to 2500-odd up near Perth, meaning that locating the
Chicken Treat in Albany just by street name is subject to a considerable
margin of error. Assuming that you'd ever want to locate a Chicken Treat.
As Martijn van Exel comments, a good approach is to have start/end numbers
per segment (or per block in cities) so that it's at least possible to get
close to a number. Gathering number data at the junctions of streets is a
darn sight quicker than doing a whole street, and gives a good gain in
address accuracy for a smaller investment of time. I managed to do junction
numbers for a large chunk of the suburb I live in in only about twice the
time it took to do the street I live on (having to check numbers for
subdivided plots slows things down a lot).
> OSM already has plenty of tools for 'noname' hunting but it is harder to
> down streets which are missing from the map altogether.
Another problem which crops up in WA, especially at the fringes of the metro
area where there's lots of house-building going on; new streets appear
almost without warning. This is something that could be assisted with
processing of satellite or aerial images to identify roads - where there
appears to be a road and there's no corresponding road in your database,
that's where you concentrate your checking.
Turn restrictions are also hard to survey manually. A mapper on foot or
> might not pay much attention to them, and again, it is hard to know when
> you have
> all of them. They might possibly be suggested from analysis of GPS traces,
> provided we have a large number of traces for an area and they are clearly
> tagged to show which ones are for travelling by car. This is one reason
> why a
> standard tagging scheme for GPS traces is needed.
Or some of them can be spotted if you have decent resolution images to
check, for example:
if you can find an image where the cars don't obscure the arrows on
the road). If I were Google, I'd have been capturing road markings and
signs as part of Street View for eventual extraction of this sort of data
from day one:
Getting back to the question of metrics; as a manager in my day job I've had
to deal with measuring odd stuff for years - in the case of maps, I'd
suggest following a Tom Gilb-esque approach and track the number of error
reports that turn out be valid, by area, corrected for map density. In
other words, one measures how accurate the map is in use. Unfortunately, I
doubt that those sorts of figures would be released by a commercial map data
Finally (and echoing SteveC again), I'd give a +1 to the ability to do
geocoding (both directions) as the #1 difference between current open
mapping projects and commercial data. But routing (by which I mean full
endpoint-to-endpoint connected routing, cross-border, including public
transport and other non-road links) would be a pretty close second.
(speaking only for myself, not nearmap)
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the talk