Hi Daniel,
Thank you for all the work you've put into this. I'd like to offer a
couple suggestions and/or clarifications for your proposed import
process, overview though it is.
First, I think it is very important that a tasking manager is set up on
a city/by city basis only, and that only AFTER consensus is achieved
that the import should proceed in that area. I would really like to
avoid seeing the massive nationwide tasking that was set up the first
time around. We should be making it hard for people to go rogue in
regions where consensus for an import doesn't (yet) exist.
Related to this, though important enough to be a second point in it's
own right, the tasking squares need to be small enough that a single
user can manage them and inspect every single building in a task. The
first round of import used task squares that were massive, and which
couldn't be divided any further past a certain point. Even in rural
areas, it is likely inappropriate to import areas larger than 1km^2. In
central Toronto it would be (and was) idiotic. An import that doesn't
take local scale into account shouldn't proceed. "Too big to load into
JOSM" is about 100x too big to import in my opinion and is not a good
enough benchmark for import batch sizing.
That is, each import needs to be local, and not just in a superficial
sense.
I'll also add that the issue of conflation doesn't seem to have been
worked out yet except to note that it is an issue. What will we do with
the millions of buildings which will substantially overlap/duplicate
existing buildings or imports? This needs to be worked out in detail
before anything starts up again.
And what needs to be done about already existing low quality imports?
It's good to acknowledge their existence, but what will be done about
them? We've set up a task to clean up some of the mess in Toronto (
http://tasks.osmcanada.ca/project/168 ) but this is only the tip of the
iceberg.
Again, I thank everyone for their time and effort on this - we can get
this done if we go slow and do it right :-)
Best,
Nate Wessel, PhD
Planner, Cartographer, Transport Nerd
NateWessel.com <https://www.natewessel.com>
On 2020-01-03 3:40 p.m., Daniel @jfd553 wrote:
Bonjour groupe, mes excuses pour ce très long courriel !-)
I have reviewed everything that has been written on the ODB import
(aka Canada Building Import) in Talk-ca and the wiki. I proposed
changes to some wiki pages (via talk tabs) to ease the discussions
about this import and the following. Now, in order to restart the
import, here are some thoughts and a proposal on how to proceed to
complete the task.
*1. Issues with the ODB Data Import*
Many concerns were raised about the import. One major concern was to
obtain local communities’ buy-in in the Canadian context. Another
concern was to improve the quality of the data prior the import. The
following paragraphs intend to clear most of these concerns.
*1.1. Which data import project?*
According to the import guidelines (steps 3 & 4), a data import
explicitly refers to a single data source (ODB in our case).
Discussions about the availability and quality of Microsoft or ESRI
data, while interesting, are not relevant as they should be dealt with
as other import projects.
*1.2. What has been imported so far?*
According to what I found [1], the ODB import is completed for 21
municipalities. These imports seem to have kept OSM content’s history,
at least for the samples checked, but many problems were found. In
some case, the imports brought swimming pools in OSM because they were
included in the dataset (e.g. Moncton). In other cases, importing
buildings with accurate locations (XY) over content mapped from less
accurate imagery resulted in buildings that now overlap the street
network (e.g. Squamish). It means that all these 21 imports need to be
carefully re-examined and corrected as required.
For 12 other municipalities, the import is partial, either suspended
as requested, or because previous imports had already provided most of
the buildings (often from the same municipal provider). That said the
import will definitely improve OSM accuracy and completeness if done
properly.
*2. How should ODB Data be imported?*
I will copy the following paragraphs in the “Canada Building Import”
wiki page [3] for a detailed discussion…
Since the data (ODB, OSM and imagery) differ from one municipality to
another, there can be no imports at the national or provincial level.
We have to work on a municipal basis and make sure to identify all the
problems and the corrective measures to apply when dealing with issues
like those I identified [1].
*2.1 Importing Locally*
According to the import guidelines (step 2), we must not import the
data without local buy-in. However, and contrarily to some European
country, there is usually no such thing as a local OSM community in
each municipality. However, we may find a few local mappers from time
to time. Working on a municipal basis should allow identifying these
local mappers before doing the import. I often use this tool [2] to
identify and contact local mappers. Once identified, I suggest that…
- We contact them to explain our intents by referring to appropriate
wiki pages.
- We wait a week or two to let them respond nothing, that they have
concerns, or wish to help.
- Without negative answers we could proceed to the import.
I first suggest that when a contributor wishes to import ODB for a
given municipality, he first identifies himself as responsible for the
import (we need to create an entry for each municipality somewhere in
the wiki). He can then contact local mappers, as explain above, and go
ahead with the import once everything settled. For those who already
made the import, I suggest that they review their work since many
issues were detected with some of these imports.
Since there are only a few local OSM communities in Canada, and
because Canada is large, I suggest not limiting the import of a given
municipality to the people of the concerned province or region.
*2.2 Pre-processing*
Once local mappers have agreed, some pre-processing can be done if
required.
A few months ago, I developed a tool that could be used to process the
data [4]. Concerns were raised because the application was developed
using proprietary software. So I documented the whole process and
algorithms in order to see courageous coders converting it in open
source software. In the meantime, and as long as I have access to an
FME licence, I could process the data, when necessary, prior to make
it available through the task manager.
Proposed pre-processing [4] includes:
- Reading of original ODB data,
- Removal of near collinear nodes (simplification),
- Orthogonalization of buildings (for corners having near right angles),
- Tagging of building footprints,
- Providing files in OSM format.
/Proposed tagging:/ In addition to the tags produced by the
orthogonalization process [4] and the source tag (source
<https://wiki.openstreetmap.org/wiki/Key:source>=Statistics Canada -
Open Building Database), the name of the Census Subdivision provided
in ODB data [5] is used to add the addr:city tag to each building.
The pre-processing requires parameters that are specific to the data
to process. These parameters were estimated on a municipal basis using
actual ODB data. The processing time increases exponentially according
to the number of buildings so, it may take a couple of days before the
data is available for a given municipality. Currently, the proposed
pre-processing does not convert terrace buildings into individual
houses nor it tags topological errors.
*2.3. Import Process*
After the local mappers, if any, agreed to the import, the
pre-processing completed when required, we can proceed to the import.
1- Do not bulk import the data! Always use the task manager
(http://tasks.osmcanada.ca/). Select and open a task square in JOSM.
If it’s too big (e.g. too much work or request is too big to load in
JOSM), go back to the task manager and split the task into smaller
squares.
2- Load imagery layer (Bing or ESRI World Imagery) and align the
imagery with ODB data (i.e. create a new image offset) if necessary
because, unless proven otherwise, ODB should be more accurate (XY)
than most available images especially in hilly areas.
3- Align the existing OSM content to the image (i.e. after the new
offset is applied) if required.
4- Currently step 2 and following as described in the wiki [2]. I
suggest merging the Conflation section [6] here and reviewing
everything to take into account the current proposal.
*References*
[1] https://wiki.openstreetmap.org/wiki/The_Open_Database_of_Buildings__
[2]
https://wiki.openstreetmap.org/wiki/Canada_Building_Import#Import_process
[3] http://resultmaps.neis-one.org(“Overview of OpenStreetMap
Contributors aka who’s around me?”)
[4] https://github.com/jfd553/OrthogonalizingBuildingFootprint__
[5]
https://www150.statcan.gc.ca/n1/pub/92-195-x/2011001/geo/csd-sdr/csd-sdr-eng.htm
[6] https://wiki.openstreetmap.org/wiki/Canada_Building_Import#Conflation
Let’s move ahead!
Daniel
_______________________________________________
Talk-ca mailing list
[email protected]
https://lists.openstreetmap.org/listinfo/talk-ca
_______________________________________________
Talk-ca mailing list
[email protected]
https://lists.openstreetmap.org/listinfo/talk-ca
