[Tagging] POI data and Addresses on areas - Was: addresses on buildings

Jarek Piórkowski jarek at piorkowski.ca
Sat Jan 11 02:39:44 UTC 2020


On Fri, 10 Jan 2020 at 18:04, Florian Lohoff <f at zz.de> wrote:
> On Fri, Jan 10, 2020 at 09:34:32AM -0500, Jarek Piórkowski wrote:
> > On Fri, 10 Jan 2020 at 04:22, Florian Lohoff <f at zz.de> wrote:
> > > OTOH in the dense urban areas you have the problem of Address for road A
> > > nearer to Road B. So you get navigated to the wrong spot on the road
> > > network. This view is generated with the OSRM Car profile and mapping
> > > all addr:* objects with the "nearest" function and comparing the highway
> > > name and the addr:street. If both are "filled" and non equal -> fail.
> > >
> > > https://osm.zz.de/dbview/?db=addresses-owl&layer=namemismatch#51.98848,8.49342,18z
> >
> > In the case of Cheruskerstraße 107g it looks like by far the easiest
> > way to solve it is for the router to take footways into account. Or I
> > guess we could create a new tag for "motor vehicle stop location to
> > get to a given address" to work around router shortcomings...?
>
> How can a router take footways into account when your mode of transport
> is by car? Can it take ALL footways in the routing graph? Only some?
> Which?

I guess it could be only those footways that change which motor
vehicle road is nearest to a given destination. That is, ignore most
sidewalks.

--

I was thinking about this whole thing earlier. Caution, wall of text.

At the risk of being philosophical, what is an address exactly?

Our wiki doesn't specify which address we're talking about:

1. Where lettermail is delivered?
2. Where a package or a shipment of a given size is delivered?
3. Where a vehicle of a given type making the delivery would stop?
4. Where a pizza is delivered?
5. Where a vehicle of a given type making the pizza delivery would stop?
6. Where emergency responders go?
7. What a government says?

In the specific case of Cheruskerstraße 107g in Bielefeld given above,
I'm guessing (with some familiarity with German urban built form, but
not the specific area): #2, #4, #6 are a front door of the building
tagged with the address; #1 might be away from that front door; #3 and
#5 are probably along Cheruskerstr. about 80 m away or maybe in the
little parking lot near the intersection with Auf den Köppen; #7 is
unknown - if that complex is a condominium or similar structure, 107g
might not even exist as far as land registry is concerned.

In some regions, land plots are identified by addresses (that is,
owners are registered with a cadastral authority or a land registry as
owning a piece of land identified as 123 Main Street). Given enough
data with a suitable licence, we could have each land plot as a way
and tag it with the address. These land plots might have zero, one, or
more buildings on it. In some regions, some of these buildings are
commonly identified as having the same address, some (like garages or
sheds) usually not. The boundaries of the building are often more
verifiable than the boundaries of the land plot, because they are more
reliably seen on satellite imagery.

In some regions, cadastral registries do not identify plots with
addresses but with other schemes.

In another definition, an address is somewhere where the local mail
service (or sometimes competitive mail services) delivers lettermail.
In some regions this is usually near a building, a mailbox attached to
the building itself, or a mail slot in the building door. In some
other regions, this is usually a mailbox near the nearest public road
to the building, and some yet other regions have community mailboxes,
so the lettermail or small package delivery location is quite far away
from where the people receiving this mail actually live.

Some buildings have more than one postal address. I suppose there are
also cases of buildings that occupy more than one legal land plot. And
of course there are addresses that might have mailboxes or accept
deliveries but not actually have a building associated with that
address.

In cases where land plots are identified by addresses, the cadastral
address might not match the postal address (due to drift of
definitions over time like the London post town or the Ontario
municipal amalgamations, or just straight-up database errors, or
inconsistencies: Canada Post and City of Toronto don't agree on
spelling of several street names).

There are also deliveries, such as packages, larger shipments, or food
delivery. They don't necessarily go the same location as the
lettermail mailbox. In some cases a motor vehicle will stop as close
as you can to the door and deliver; in other cases there are explicit
systems of driveways and loading docks often quite far away from main
entrances or mailboxes. In some cases, the locally prescribed way of
making a delivery is to use a designated delivery parking area and
deliver the last meters on foot or with a cargo trolley. You might
want to know both where to park a vehicle and where the receiver's
door is. Then there's increasingly different kinds of delivery
vehicles, some with different legal access rights ("zero emission
vehicles only"), some with different physical dimensions (bicycle
couriers), in many cases both, so each might need a separate
delivery-stopping area.

Emergency services may also use addresses to direct responders. Source
of these addresses is usually either cadastral or postal, sometimes
other. In some cases emergency responders might define a location by a
reference to a street address but accessed in a substantially
different way ("in the alley behind 640 Union Street", e.g.
http://archive.is/X6zQR ).

I suppose there are some cases where locations are identified by a
commonly known address but none of the above actually apply (does
Royal Mail deliver to 10 Downing Street? maybe?). I guess we can
ignore made-up but functioning postal addresses like "Santa Claus,
North Pole, H0H0H0".

And that's before getting into locations that don't have a cadastral
authority, a concept of land or building ownership, or a public postal
service.

In terms of surveying and verifiability, basically the only thing
surveyable is an address displayed on the property by its owner or
operator, whether on a building, on a mailbox, or an address tag on a
plot of land. For buildings, bounds can often be estimated from
streetside survey coupled with aerial imagery (but not always, as in
the case of complex buildings); for land, bounds are often not easily
verifiable. Where suitably-licensed datasets exist, they could be used
to add in boundaries or unsigned addresses at cost of surveyability.
Where displayed by owner, addresses will sometimes be on a building
(thus identifying at least one building with that address), sometimes
on a mailbox, sometimes both. Sometimes a mailbox address strongly
implies a building address (as in the case of a mailbox at the bottom
of a driveway leading to only one building), sometimes it does not.
Unless signed explicitly ("deliveries for 333 Bay Street"), parking
and stopping locations might not be easily verifiable.

Going back to the campsite-across-the-river problem:

> The processing of finding a point on the routeable network is a spatial
> "nearest". I know if no nav solution which does something more
> sophisticated. I'd like to be proven wrong.

I am not quite sure how we could actually resolve the problem of
routers not taking spatial barriers into account other than by either
mapping an address according to definition #3 (parking) rather than #6
(emergency), or as a series of waypoints. We could have a node for
"entrance leading to point XYZ" _on_ the routable network. But then
how do you actually get to the destination front door, or the campsite
reception, or your camping spot along these access=private roads or
footways - switch router mode once past the entrance node? To fully
describe a delivery you might need a node for "route here for
deliveries to XYZ" on the public car routable network, then a ferry
with motor_vehicle=private (or if we're lucky it's got delivery=yes),
another roadway connecting to a "enter here for deliveries to XYZ",
and connecting to that a private footway (and maybe some elevators)
leading to the door. Are these waypoints verifiable?


So, are the seven points above different kinds of addresses? How do we
tag these all? Or are some of them not addresses? Or how do we choose
which ones _not_ to tag? What other kinds of addresses have I missed?

--Jarek



More information about the Tagging mailing list