Re: [Talk-ca] using image recognition to create building foot prints.

[Matthew Darwin]

Ok, just I had to re-write more than just that one section to make it 
coherent.   Hopefully the page has a bit more logical structure now.


https://wiki.openstreetmap.org/wiki/WikiProject_Canada/Building_Canada_2020

Comments and edits welcome!


On 2018-01-29 09:26 PM, Matthew Darwin wrote:


I will re-write the "project management" section of the wiki page to 
align with this discussion.


On 2018-01-29 09:17 PM, john whelan wrote:
and I think I agree with Pierre the best approach would be to do it 
a step at a time using experienced local resources.


We do need to engage with high schools and the Universities but it 
is difficult with the resources available.  There is some material 
available https://wiki.openstreetmap.org/wiki/Education but it 
would need reviewing to see if it is relevant to what is required.


We were exceptionally fortunate in Ottawa with the pilot and the 
resources we had available but even there I wasn't certain we would 
be able to pull it off.


Cheerio John


2018-01-29 21:11 GMT-05:00 Matthew Darwin >:


+1

Unless someone has lots of $$$ to throw at OSM work (which
could then fund full time coordinators, trainers, lawyers,
etc), the only way I see to coordinate is to approach it like
how OSM in Canada was build up to now... distributed model with
local groups doing what makes sense for their area.   There is,
of course, no way to map all buildings in Canada by 2020 this
way.  Still it is good to set aspirational goals...

On 2018-01-29 08:57 PM, Pierre Béland wrote:


Il faut une part de réalisme. Pour bien coordonner, il ne
suffit pas de créer une tâche et d'inviter à participer. Nous
ne sommes pas une communauté structurée au niveau national. 
Je comprends que diverses universités s'intéressent au projet
OSM et aimeraient initier leurs étudiants à ce projet. La
meilleure solution je pense c'est de se mettre en contact avec
la communauté OSM locale et s'assurer de bien encadrer la
formation et les premiers jours de participation à OSM.

Les contributeurs sont davantage actifs dans leurs communautés
locales ou selon leur divers intérêts liés à leur travail ou
loisir. Personne n'est prêt à s'engager à coordonner un tel
projet au niveau national.



___
Talk-ca mailing list
Talk-ca@openstreetmap.org 
https://lists.openstreetmap.org/listinfo/talk-ca







___
Talk-ca mailing list
Talk-ca@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-ca

Re: [Talk-ca] using image recognition to create building foot prints.

[Matthew Darwin]

I will re-write the "project management" section of the wiki page to 
align with this discussion.


On 2018-01-29 09:17 PM, john whelan wrote:
and I think I agree with Pierre the best approach would be to do it 
a step at a time using experienced local resources.


We do need to engage with high schools and the Universities but it 
is difficult with the resources available.  There is some material 
available https://wiki.openstreetmap.org/wiki/Education but it would 
need reviewing to see if it is relevant to what is required.


We were exceptionally fortunate in Ottawa with the pilot and the 
resources we had available but even there I wasn't certain we would 
be able to pull it off.


Cheerio John


2018-01-29 21:11 GMT-05:00 Matthew Darwin >:


+1

Unless someone has lots of $$$ to throw at OSM work (which could
then fund full time coordinators, trainers, lawyers, etc), the
only way I see to coordinate is to approach it like how OSM in
Canada was build up to now... distributed model with local
groups doing what makes sense for their area.   There is, of
course, no way to map all buildings in Canada by 2020 this way.
Still it is good to set aspirational goals...

On 2018-01-29 08:57 PM, Pierre Béland wrote:


Il faut une part de réalisme. Pour bien coordonner, il ne
suffit pas de créer une tâche et d'inviter à participer. Nous
ne sommes pas une communauté structurée au niveau national.  Je
comprends que diverses universités s'intéressent au projet OSM
et aimeraient initier leurs étudiants à ce projet. La meilleure
solution je pense c'est de se mettre en contact avec la
communauté OSM locale et s'assurer de bien encadrer la
formation et les premiers jours de participation à OSM.

Les contributeurs sont davantage actifs dans leurs communautés
locales ou selon leur divers intérêts liés à leur travail ou
loisir. Personne n'est prêt à s'engager à coordonner un tel
projet au niveau national.



___
Talk-ca mailing list
Talk-ca@openstreetmap.org 
https://lists.openstreetmap.org/listinfo/talk-ca





___
Talk-ca mailing list
Talk-ca@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-ca

Re: [Talk-ca] using image recognition to create building foot prints.

[john whelan]

 and I think I agree with Pierre the best approach would be to do it a step
at a time using experienced local resources.

We do need to engage with high schools and the Universities but it is
difficult with the resources available.  There is some material available
https://wiki.openstreetmap.org/wiki/Education but it would need reviewing
to see if it is relevant to what is required.

We were exceptionally fortunate in Ottawa with the pilot and the resources
we had available but even there I wasn't certain we would be able to pull
it off.

Cheerio John


2018-01-29 21:11 GMT-05:00 Matthew Darwin :

> +1
>
> Unless someone has lots of $$$ to throw at OSM work (which could then fund
> full time coordinators, trainers, lawyers, etc), the only way I see to
> coordinate is to approach it like how OSM in Canada was build up to now...
> distributed model with local groups doing what makes sense for their
> area.   There is, of course, no way to map all buildings in Canada by 2020
> this way.  Still it is good to set aspirational goals...
> On 2018-01-29 08:57 PM, Pierre Béland wrote:
>
>
> Il faut une part de réalisme. Pour bien coordonner, il ne suffit pas de
> créer une tâche et d'inviter à participer. Nous ne sommes pas une
> communauté structurée au niveau national.  Je comprends que diverses
> universités s'intéressent au projet OSM et aimeraient initier leurs
> étudiants à ce projet. La meilleure solution je pense c'est de se mettre en
> contact avec la communauté OSM locale et s'assurer de bien encadrer la
> formation et les premiers jours de participation à OSM.
>
> Les contributeurs sont davantage actifs dans leurs communautés locales ou
> selon leur divers intérêts liés à leur travail ou loisir. Personne n'est
> prêt à s'engager à coordonner un tel projet au niveau national.
>
>
>
> ___
> Talk-ca mailing list
> Talk-ca@openstreetmap.org
> https://lists.openstreetmap.org/listinfo/talk-ca
>
>
___
Talk-ca mailing list
Talk-ca@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-ca

Re: [Talk-ca] using image recognition to create building foot prints.

[Matthew Darwin]

+1

Unless someone has lots of $$$ to throw at OSM work (which could then 
fund full time coordinators, trainers, lawyers, etc), the only way I 
see to coordinate is to approach it like how OSM in Canada was build 
up to now... distributed model with local groups doing what makes 
sense for their area.   There is, of course, no way to map all 
buildings in Canada by 2020 this way.  Still it is good to set 
aspirational goals...


On 2018-01-29 08:57 PM, Pierre Béland wrote:


Il faut une part de réalisme. Pour bien coordonner, il ne suffit pas 
de créer une tâche et d'inviter à participer. Nous ne sommes pas une 
communauté structurée au niveau national.  Je comprends que diverses 
universités s'intéressent au projet OSM et aimeraient initier leurs 
étudiants à ce projet. La meilleure solution je pense c'est de se 
mettre en contact avec la communauté OSM locale et s'assurer de bien 
encadrer la formation et les premiers jours de participation à OSM.


Les contributeurs sont davantage actifs dans leurs communautés 
locales ou selon leur divers intérêts liés à leur travail ou loisir. 
Personne n'est prêt à s'engager à coordonner un tel projet au niveau 
national.


___
Talk-ca mailing list
Talk-ca@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-ca

Re: [Talk-ca] using image recognition to create building foot prints.

[Pierre Béland]

Effectivement, on en revient aux discussions d'il y a quelques mois. Nous avons 
indiqué dès le départ que ce projet était très ambitieux et qu'il y avait des 
risques importants sur la qualité de la données si on s'appuyait principalement 
sur des groupes inexpérimentés via les universités, etc. 
John mentionnait la réponse pour le Népal. Nous venions de terminer la réponse 
pour l'Ébola après près d'un an de cartographie intensive.  Nous avions 
plusieurs défis à surmonter avec l'absence d'imagerie récente, des communautés 
isolées en montagne, des communications interrompues.  A cela s'est ajoutée la 
vague déferlante des carto-parties. Les médias avaient beaucoup parlé de la 
réponse humanitaire OSM aux diverses crises humanitaires. Beaucoup de groupes 
étaient intéressés à collaborer. Encore plus que pour Tacloban aux Philippines 
et pour l'Ebola.  Les statistiques montraient des records de participation de 
nouveaux contributeurs. En contrepartie, John et d'autres qui validaient les 
données, rapportaient continuellement des problèmes de qualité de la donnée. 
Nous avons vécu le même problème avec Haiti en octobre 2016. 

Il faut une part de réalisme. Pour bien coordonner, il ne suffit pas de créer 
une tâche et d'inviter à participer. Nous ne sommes pas une communauté 
structurée au niveau national.  Je comprends que diverses universités 
s'intéressent au projet OSM et aimeraient initier leurs étudiants à ce projet. 
La meilleure solution je pense c'est de se mettre en contact avec la communauté 
OSM locale et s'assurer de bien encadrer la formation et les premiers jours de 
participation à OSM.

Les contributeurs sont davantage actifs dans leurs communautés locales ou selon 
leur divers intérêts liés à leur travail ou loisir. Personne n'est prêt à 
s'engager à coordonner un tel projet au niveau national.  
Des personnes comme John et moi avons consacré beaucoup de temps à coordonner, 
supporter les réponses humanitaires OSM au niveau international. Nos propos 
prudents visent à en venir à une approche réaliste.  Il vaut mieux partir sur 
des bases solides, initier quelques projets au départ si des communautés sont 
prêtes à les supporter.    
Pierre 
 

Le lundi 29 janvier 2018 18:57:14 HNE, john whelan  
a écrit :  
 
 ​The idea behind the building project is good.  Basically you need a mixture 
of accurate building outlines and tags.  From there the statisticians can work 
their magic.  This is true in Canada as well as other parts of the world.  With 
OSM buildings and tags combined with open source stats software R (R.org) you 
get ground floor GIS planning tools and they are badly needed in Africa etc.  
If we can pull it off this is good.

First Pierre's point that machine learning and imagery is a little different to 
using radar type technology. If we would have had it available in Nepal it 
would have saved a lot of problems.  Even today in Africa the standard of 
building mapping would be much improved by the use of such technology.

It isn't yet accepted by OSM as mainstream.  That is an issue we need to get 
round before we can use the stuff.

However Canada has always been in the forefront of imports.  We have a history 
of using NCR Canada’s data ie CANVEC and we are comfortable using it.  Some 
parts we recognise are better quality than others.  We also have within the 
mailing list some deep technical expertise which can be used to evaluate the 
radar type technology for detecting building outlines.  I think it will take 
time to get this technology accepted by OSM and that is the point of this 
thread.

I think we have to accept that the BC2020i project is one that was not dreamt 
up by the OSM community.  I think the idea came out of Alessandro of Stats 
Canada and my understanding is the web page was put together by a single person 
with little experience of OSM, the processes and politics involved. There is 
demand for the data but OSM is more geared towards mappers than customer 
demands.

What we have ended up with is a project with lots of words and aspirations but 
little apparent understanding of what is involved.  The idea has been picked up 
by High Schools and Universities and we are now getting inexperienced mappers 
in with little training adding buildings to the map in iD and the data quality 
is poor for some and that is an issue since it reflects on the project itself.

There seems to be no project manager and that is an issue.  We’ve cleaned up 
the wiki page to some extent.  There is a demand from schools and Universities 
to get involved.  We need someone to put together guidance for these people.

I take Pierre’s point that in an ideal world experienced local mappers would 
map locally and take responsibility for their area importing when appropriate.  
Unfortunately we do not live in an ideal world.

Cheerio John​

On 29 January 2018 at 17:59, OSM Volunteer stevea  
wrote:

On Jan 29, 2018, at 2:35 PM, 

Re: [Talk-ca] using image recognition to create building foot prints.

[john whelan]

​The idea behind the building project is good.  Basically you need a
mixture of accurate building outlines and tags.  From there the
statisticians can work their magic.  This is true in Canada as well as
other parts of the world.  With OSM buildings and tags combined with open
source stats software R (R.org) you get ground floor GIS planning tools and
they are badly needed in Africa etc.  If we can pull it off this is good.

First Pierre's point that machine learning and imagery is a little
different to using radar type technology. If we would have had it available
in Nepal it would have saved a lot of problems.  Even today in Africa the
standard of building mapping would be much improved by the use of such
technology.

It isn't yet accepted by OSM as mainstream.  That is an issue we need to
get round before we can use the stuff.

However Canada has always been in the forefront of imports.  We have a
history of using NCR Canada’s data ie CANVEC and we are comfortable using
it.  Some parts we recognise are better quality than others.  We also have
within the mailing list some deep technical expertise which can be used to
evaluate the radar type technology for detecting building outlines.  I
think it will take time to get this technology accepted by OSM and that is
the point of this thread.

I think we have to accept that the BC2020i project is one that was not
dreamt up by the OSM community.  I think the idea came out of Alessandro of
Stats Canada and my understanding is the web page was put together by a
single person with little experience of OSM, the processes and politics
involved. There is demand for the data but OSM is more geared towards
mappers than customer demands.

What we have ended up with is a project with lots of words and aspirations
but little apparent understanding of what is involved.  The idea has been
picked up by High Schools and Universities and we are now getting
inexperienced mappers in with little training adding buildings to the map
in iD and the data quality is poor for some and that is an issue since it
reflects on the project itself.

There seems to be no project manager and that is an issue.  We’ve cleaned
up the wiki page to some extent.  There is a demand from schools and
Universities to get involved.  We need someone to put together guidance for
these people.

I take Pierre’s point that in an ideal world experienced local mappers
would map locally and take responsibility for their area importing when
appropriate.  Unfortunately we do not live in an ideal world.

Cheerio John​


On 29 January 2018 at 17:59, OSM Volunteer stevea  wrote:

> On Jan 29, 2018, at 2:35 PM, Stewart C. Russell  wrote:
> > On 2018-01-29 04:37 PM, OSM Volunteer stevea wrote:
> >>
> >> OSM is delighted to receive building data in Canada, truly we are.
> >> (Provided they are high-quality data).  I have heard the process of
> >> entering data into OSM, especially "bulk import" OD (which must match
> >> license compatibility against OSM's license, our ODbL) described as
> >> "inside baseball."  It is not.
> >
> > If you're gonna quote me, at least try to understand me, please.
>
> Stewart, I apologize if I misquoted you or took it out of context, that
> was not my intention.
>
> > The open data / OSM dialogue in Canada has been going something like
> > this, ever since I started working with municipal groups in 2011 or so:
> >
> > Municipal data advocate: Please use our data! It's under an open
> >   licence!
> >
> > OSM volunteer: But our licences aren't compatible!
> >
> > Municipal data advocate: But it's an open licence! Our lawyers say
> >   you'll be fine!
> >
> > OSM volunteer: But we need … (starts to reel off list of additional
> >   supporting docs)
> >
> > Municipal data advocate: Companies like Google and Nokia use our data
> >   with no problem. Use our data! We are giving it to you!
> >   Don't complain!
> >
> > OSM volunteer: but but the licence …
> >
> > (Municipal data advocate storms off in search of a someone more likely
> > to give them corporate recognition.)
> >
> > Some very tenacious OSM people and some very adaptable government people
> > have made things work in a few places in Canada.
>
> I both salute these efforts and bow deeply in obeisance at the good work
> done here by them.  These are the important seeds of the future, the acorns
> from which mighty oaks shall grow.  Yes, it will take time, effort,
> coordination, management and documentation.
>
> Simply put, (and I don't wish to be rude), "municipal data advocates"
> cannot assume that OSM is a "free ride," without some front-loaded effort
> at planning and further project guidance along the way.  We have our
> culture and methods in OSM, and that's the way it is.  I am hopeful, as
> inter-community cooperation is something Canada has been and is quite good
> at doing for its entire history.
>
> > Only when we have a way
> > forward on data licensing, 

Re: [Talk-ca] using image recognition to create building foot prints.

[OSM Volunteer stevea]

On Jan 29, 2018, at 2:35 PM, Stewart C. Russell  wrote:
> On 2018-01-29 04:37 PM, OSM Volunteer stevea wrote:
>> 
>> OSM is delighted to receive building data in Canada, truly we are.
>> (Provided they are high-quality data).  I have heard the process of
>> entering data into OSM, especially "bulk import" OD (which must match
>> license compatibility against OSM's license, our ODbL) described as
>> "inside baseball."  It is not.
> 
> If you're gonna quote me, at least try to understand me, please.

Stewart, I apologize if I misquoted you or took it out of context, that was not 
my intention.

> The open data / OSM dialogue in Canada has been going something like
> this, ever since I started working with municipal groups in 2011 or so:
> 
> Municipal data advocate: Please use our data! It's under an open
>   licence!
> 
> OSM volunteer: But our licences aren't compatible!
> 
> Municipal data advocate: But it's an open licence! Our lawyers say
>   you'll be fine!
> 
> OSM volunteer: But we need … (starts to reel off list of additional
>   supporting docs)
> 
> Municipal data advocate: Companies like Google and Nokia use our data
>   with no problem. Use our data! We are giving it to you!
>   Don't complain!
> 
> OSM volunteer: but but the licence …
> 
> (Municipal data advocate storms off in search of a someone more likely
> to give them corporate recognition.)
> 
> Some very tenacious OSM people and some very adaptable government people
> have made things work in a few places in Canada.

I both salute these efforts and bow deeply in obeisance at the good work done 
here by them.  These are the important seeds of the future, the acorns from 
which mighty oaks shall grow.  Yes, it will take time, effort, coordination, 
management and documentation.

Simply put, (and I don't wish to be rude), "municipal data advocates" cannot 
assume that OSM is a "free ride," without some front-loaded effort at planning 
and further project guidance along the way.  We have our culture and methods in 
OSM, and that's the way it is.  I am hopeful, as inter-community cooperation is 
something Canada has been and is quite good at doing for its entire history.

> Only when we have a way
> forward on data licensing, then BC2020 would be an OSM project.

I respectfully disagree.  Yes, "ways forward on data licensing" is vitally 
important, as it is a major obstacle.  However, BC2020, and the way that it has 
morphed into becoming by its very nature using OSM as a repository of data, IS 
an OSM project.  Therefore, it must hew to OSM tenets, like transparency, good 
communication, wiki updates, and in a project of scope this wide, sane and 
steady planning and project management.

You can say that license compatibility is "slow going" (and you'd be right) but 
OSM is "up to" three cities (from one, Ottawa).  Rome wasn't built in a day and 
Canada's building data won't be entered into OSM in a day, either.  HOWEVER, as 
they are being entered now, some "manually," some (few) via OD licence, and 
some as simple improvements ("Hey, I'm going to tag this a café because I'm a 
local OSM user and I know it is one!"), these efforts MUST BE coordinated (or 
managed, I keep saying, though I'm not particularly enamored of the word as it 
seems non-OSM, yet on national-scope projects, something like "management" 
really is required, even if it is "loose but effective coordination").

Building data being entered into OSM do not have to be part of BC2020i, what is 
now WikiProject BC2020:  if I simply tag a building polygon amenity=cafe, I 
don't become part of a coordinated effort.  However, to the extent they strive 
to be part of the coordinated effort to enter nationwide building data, 
following the guidelines in our wiki of what we mean by acceptable-quality 
data, with acceptable tags, they really, really should.  Such coordination 
benefits everybody, and at minor "cost" (follow some nationwide guidelines, 
stay communicative with your status...).  Is that so difficult a point upon 
which to agree?

Thank you for continuing good dialog,
SteveA
___
Talk-ca mailing list
Talk-ca@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-ca

Re: [Talk-ca] using image recognition to create building foot prints.

[Stewart C. Russell]

On 2018-01-29 04:37 PM, OSM Volunteer stevea wrote:
> 
> OSM is delighted to receive building data in Canada, truly we are.
> (Provided they are high-quality data).  I have heard the process of
> entering data into OSM, especially "bulk import" OD (which must match
> license compatibility against OSM's license, our ODbL) described as
> "inside baseball."  It is not.

If you're gonna quote me, at least try to understand me, please.

The open data / OSM dialogue in Canada has been going something like
this, ever since I started working with municipal groups in 2011 or so:

Municipal data advocate: Please use our data! It's under an open
licence!

OSM volunteer: But our licences aren't compatible!

Municipal data advocate: But it's an open licence! Our lawyers say
you'll be fine!

OSM volunteer: But we need … (starts to reel off list of additional
supporting docs)

Municipal data advocate: Companies like Google and Nokia use our data
with no problem. Use our data! We are giving it to you!
Don't complain!

OSM volunteer: but but the licence …

(Municipal data advocate storms off in search of a someone more likely
to give them corporate recognition.)

Some very tenacious OSM people and some very adaptable government people
have made things work in a few places in Canada. Only when we have a way
forward on data licensing, then BC2020 would be an OSM project.

 Stewart

 Stewart

Stewart

___
Talk-ca mailing list
Talk-ca@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-ca

Re: [Talk-ca] using image recognition to create building foot prints.

[Pierre Béland]

Précision,   Les missions aériennes permettent de produire des images de grande 
qualité.  On y associe des équipements LIDAR qui émettent un signal vers le sol 
pour mesurer la distance. Aussi bien la technique LIDAR que de petits drones 
sont aujourd'hui capables de produire des modèles d'élévation avec quelques 
centimètres de précision.  Cela permet aussi de produire des modèles 3D des 
immeubles et de distinguer avec la végétation.
Suite aux inondations du Richelieu et du Lac Champlain en 2011, des modèles 
d'élévation très des zones urbaines en bordure de la rivière Richelieu ont été 
produites. Si on se rappelle les discussions il y a quelques mois, un tel 
travail d'import va nécessiter des ressources importantes. Les diverses 
communautés OSM locales devront évaluer leur capacité à réaliser des projets 
d'import d'immeubles. Et il faut éviter de se baser sur le modèle 
«Cartoparties» pour réaliser de tels projets. Des milliers de personnes qui 
sont sensibiliées quelques heures à la cartographie ne reviennent pas ensuite 
et laisse souvent une donnée de piètre qualité.
Il faut être réaliste et construire peu à peu, motiver des communautés locales 
à expérimenter un modèle d'import de la donnée. Cela fera ensuite boule de 
neige ( c'est de saison :)  )

Pierre 
 

Le lundi 29 janvier 2018 16:07:31 HNE, Pierre Béland  a 
écrit :  
 
 Bonjour John
Les spécialistes d'imagerie produisent des couches de données assez précises à 
partir d'imagerie LIDAR ou de drones, incluant, immeubles, cours d'eau et 
occupation du sol. Ces images offrent qualité et précision. Des techniques de 
classification, interprétation, correction permettent aux spécialistes de 
converger vers un produit de qualité.  Et bien sûr toutes ces avancées 
technologiques et l'accès éventuel à des profanes bousculent les habitudes tout 
comme les véhicules sans conducteur :)
Même si Statistique Canada fournit à OSM un fichier produit par des 
spécialistes, il sera nécessaire ensuite d'établir une procédure d'import, de 
fusionner / aligner avec les données existantes et de corriger. 
Et ouvront la porte au Futur! Une autre avenue, c'est l'accès aux profanes que 
nous sommes à des outils semi-automatiques pour faciliter la digitalisation de 
différents éléments tels immeubles, routes et rivières. Je ne connais pas 
l'historique des expériences d'utilisation de tels outils. Mais on peut 
remonter en 2011, où on parlait d'un outil de détection de route. Divers 
articles traitent aussi de ce sujet.
https://alastaira.wordpress.com/2011/02/04/automatic-road-detection-using-bing-maps-imagery/https://gis.stackexchange.com/questions/77876/is-there-a-tool-that-performs-automatic-recognition-of-buildings


Facebook a aussi expérimenté des outils de reconnaissance d'image en Thaîlande 
récemment. Selon les plaintes de certains contributeurs, les données ont été 
ajoutées à OSM sans valider suffisamment avec la réalité sur le terrain.
Je pense qu'il serait intéressant pour les contributeurs expérimentés d'avoir 
accès à des outils semi-automatisés facilitant dans JOSM par exemple le tracé 
d'immeubles, routes, cours d'eau, etc. Pour un cours d'eau par exemple, je 
déplace le curseur de la souris, et les contours et le centre de la rivière 
sont tracés automatiquement. Ou encore le contour d'un lac est tracé.
 
Pierre 
 

Le lundi 29 janvier 2018 15:15:59 HNE, john whelan  
a écrit :  
 
 ·    NRCan is working on a methodology to extract building  footprints, 
including topographic elevation and height attributes, from LiDAR


Traditionally OSM has not been happy with this sort of thing.  The accuracy can 
be poor.

We probably need to think about this since the BC2020i project had this 
mentioned and Stats Can has given it a mention also.  I'm not promoting it nor 
saying its bad but it will almost certainly be raised shortly.

First if an import was done using this data who would be the local group to 
approve it?  I suspect because it covers the entire country it would be the 
talk-ca group.  The date would come through the TB portal so licensing is not 
an issue.  Or it could be split into regions with regional local groups making 
decisions.

The other very big question is to do with data quality.  So far nothing that is 
machine learnt from imagery has consistently met the expectations of 
OpenStreetMap.

Note to Pierre I'm not sure if you are on the talk-ca mailing list but any 
feedback you might have on the data quality side would be welcome and will be 
shared amongst the group.

Thoughts?

Thanks John
___
Talk-ca mailing list
Talk-ca@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-ca
___
Talk-ca mailing list
Talk-ca@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-ca

Re: [Talk-ca] using image recognition to create building foot prints.

[OSM Volunteer stevea]

Um, "dyed-in-the-wool."
Steve

___
Talk-ca mailing list
Talk-ca@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-ca

Re: [Talk-ca] using image recognition to create building foot prints.

[OSM Volunteer stevea]

On Jan 29, 2018, at 12:15 PM, john whelan  wrote:
> ·NRCan is working on a methodology to extract building footprints, 
> including topographic elevation and height attributes, from LiDAR
> Traditionally OSM has not been happy with this sort of thing.  The accuracy 
> can be poor.

I find it troublesome that John should even have to explain this to a single 
person on this list.  However, maybe some here don't know this, so "OK."

Warning:  lukewarm screed ahead.

I say this with all politeness intended:  OpenStreetMap is not a dumping 
grounds for data which are open (OD) nor experimental from AI/machine learning 
results, then there is some vague and hand-waving future process which may or 
may not come along in the future and "fix them to meet OSM's quality standards."

OSM is delighted to receive building data in Canada, truly we are.  (Provided 
they are high-quality data).  I have heard the process of entering data into 
OSM, especially "bulk import" OD (which must match license compatibility 
against OSM's license, our ODbL) described as "inside baseball."  It is not.  
Every single person reading this list is either an OSM volunteer like me, or 
has an interest in OSM.  (Signing up for this mailing list is my evidence for 
saying that).  That means fully treating OSM as the "host project" for these 
data, simply put, because it is.  Acting as if OSM is simply an airplane 
expected to fly, without following a flight plan nor contacting a control tower 
will get this project grounded, as OSM volunteers like me declare an emergency, 
seeing nobody in the pilot's seat, while all the passengers expect a free ride 
to their own distinct destinations.  That simply "doesn't fly."

Recently, many of us have become more aware of the history of StatsCan wanting 
to introduce building data to crowdsourcing and coming up with BC2020i (the i 
for "initiative").  Left quite unspoken was that "crowdsourcing building data 
from StatsCan" silently became "put building data into OSM."  While there was 
traction to do this in 2016, many meetings, a fair amount of data entry and the 
beginnings of more established OSM-style process (writing a wiki or two, 
efforts to harmonize local OD licenses with ODbL...) in 2017, this "initiative" 
has now fully morphed into an OSM WikiProject, BC2020.  This has its own wiki 
(https://wiki.openstreetmap.org/wiki/WikiProject_Canada/Building_Canada_2020 
being re-written now) and is getting on a better footing by discussing the huge 
number of moving parts both there and here in talk-ca.  But, in my opinion, the 
project remains badly unfocused (slowly, it gets better).

Building data in Canada entering OSM, whether by "manual" methods (e.g. tracing 
a Bing layer) or by "bulk import" from licenced OD is ongoing.  However, it 
absolutely must hew to the tenets of OSM:  good, wide-area coordination and 
communication, consensus, following the guidelines written into an Import Plan 
(if bulk importation of OD data happens) and keeping others informed of status 
(licence approvals, how far along manual methods are via the Task Manager...).  
This is NOT "inside baseball."  It is OSM, plain and simple, through and 
through — this is because BC2020 is an OSM project.  In my opinion (and if it 
isn't clear by now), this project suffers from a serious lack of Project 
Management.  I do not wish to chide, rather to encourage.

As OSM is "hosting" BC2020, I'd like to see more "cultural sensitivity" towards 
what OSM absolutely MUST DO on national-level projects:

• we document our intentions with a well-written wiki (due to history, 
what we have was poorly done, though it is improving),
• we publish Import Plans PER IMPORT (whether municipality, province or 
whatever),  Ottawa's was good, others must grow from this,
• we inform wider community (Canadian and worldwide OSM) with status, 
including licence and data entry progress, whether via Task Manager or 
otherwise,
• especially in early stages (and we are here now), we guide and steer 
the project towards achievable goals and realistic timeframes, doing so 
applying knowledge gained from previous (especially national-level or very wide 
scope) OSM projects.

This is not an exhaustive list.

Please, for months I've been cajoling, back-channel-communicating, 
wiki-improving and friend-making my way (OUR way) towards a simple 
understanding (and declaration?) by all involved in BC2020 that it is an OSM 
project, that it must hew closely to OSM tenets and that that is not "inside 
baseball."  OSM is simply the methods by which these Canadian building data 
find a home.  I'd like to see "local stovepipes" of sub-project data entry 
efforts and no- or little-communication turn into talk-ca threads and new wiki 
sections in our wiki, so this project becomes much more transparent and 
wide-area.  I'd especially like to see someone or some group (call it a 
"steering committee") step 

Re: [Talk-ca] using image recognition to create building foot prints.

[john whelan]

I think my concern at the moment is how to handle these in an acceptable
manner to OSM.

It would seem Stewart has some confidence in the data and I suspect you
have yourself.

So given a benediction by the mailing list to go ahead we would need to go
through the import process.

My gut feel would be some sort of manual import using JOSM one building at
a time over a trial area might be acceptable and I'm not sure if we have a
volunteer nor when the data will be available.

Cheerio John

2018-01-29 16:07 GMT-05:00 Pierre Béland :

> Bonjour John
>
> Les spécialistes d'imagerie produisent des couches de données assez
> précises à partir d'imagerie LIDAR ou de drones, incluant, immeubles, cours
> d'eau et occupation du sol. Ces images offrent qualité et précision. Des
> techniques de classification, interprétation, correction permettent aux
> spécialistes de converger vers un produit de qualité.  Et bien sûr toutes
> ces avancées technologiques et l'accès éventuel à des profanes bousculent
> les habitudes tout comme les véhicules sans conducteur :)
>
> Même si Statistique Canada fournit à OSM un fichier produit par des
> spécialistes, il sera nécessaire ensuite d'établir une procédure d'import,
> de fusionner / aligner avec les données existantes et de corriger.
>
> Et ouvront la porte au Futur! Une autre avenue, c'est l'accès aux profanes
> que nous sommes à des outils semi-automatiques pour faciliter la
> digitalisation de différents éléments tels immeubles, routes et rivières.
> Je ne connais pas l'historique des expériences d'utilisation de tels
> outils. Mais on peut remonter en 2011, où on parlait d'un outil de
> détection de route. Divers articles traitent aussi de ce sujet.
> https://alastaira.wordpress.com/2011/02/04/automatic-road-
> detection-using-bing-maps-imagery/
> https://gis.stackexchange.com/questions/77876/is-there-a-
> tool-that-performs-automatic-recognition-of-buildings
>
>
> Facebook a aussi expérimenté des outils de reconnaissance d'image en
> Thaîlande récemment. Selon les plaintes de certains contributeurs, les
> données ont été ajoutées à OSM sans valider suffisamment avec la réalité
> sur le terrain.
>
> Je pense qu'il serait intéressant pour les contributeurs expérimentés
> d'avoir accès à des outils semi-automatisés facilitant dans JOSM par
> exemple le tracé d'immeubles, routes, cours d'eau, etc. Pour un cours d'eau
> par exemple, je déplace le curseur de la souris, et les contours et le
> centre de la rivière sont tracés automatiquement. Ou encore le contour d'un
> lac est tracé.
>
>
> Pierre
>
>
> Le lundi 29 janvier 2018 15:15:59 HNE, john whelan 
> a écrit :
>
>
> ·
>
>
> *NRCan is working on a methodology to extract building footprints,
> including topographic elevation and height attributes, from LiDAR*
>
>
> *Traditionally OSM has not been happy with this sort of thing.  The
> accuracy can be poor.*
>
>
> *We probably need to think about this since the BC2020i project had this
> mentioned and Stats Can has given it a mention also.  I'm not promoting it
> nor saying its bad but it will almost certainly be raised shortly.*
>
>
> *First if an import was done using this data who would be the local group
> to approve it?  I suspect because it covers the entire country it would be
> the talk-ca group.  The date would come through the TB portal so licensing
> is not an issue.  Or it could be split into regions with regional local
> groups making decisions.*
>
>
> *The other very big question is to do with data quality.  So far nothing
> that is machine learnt from imagery has consistently met the expectations
> of OpenStreetMap.*
>
>
> *Note to Pierre I'm not sure if you are on the talk-ca mailing list but
> any feedback you might have on the data quality side would be welcome and
> will be shared amongst the group.*
>
>
> *Thoughts?*
>
> *Thanks John*
> ___
> Talk-ca mailing list
> Talk-ca@openstreetmap.org
> https://lists.openstreetmap.org/listinfo/talk-ca
>
___
Talk-ca mailing list
Talk-ca@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-ca

Re: [Talk-ca] using image recognition to create building foot prints.

[Pierre Béland]

Bonjour John
Les spécialistes d'imagerie produisent des couches de données assez précises à 
partir d'imagerie LIDAR ou de drones, incluant, immeubles, cours d'eau et 
occupation du sol. Ces images offrent qualité et précision. Des techniques de 
classification, interprétation, correction permettent aux spécialistes de 
converger vers un produit de qualité.  Et bien sûr toutes ces avancées 
technologiques et l'accès éventuel à des profanes bousculent les habitudes tout 
comme les véhicules sans conducteur :)
Même si Statistique Canada fournit à OSM un fichier produit par des 
spécialistes, il sera nécessaire ensuite d'établir une procédure d'import, de 
fusionner / aligner avec les données existantes et de corriger. 
Et ouvront la porte au Futur! Une autre avenue, c'est l'accès aux profanes que 
nous sommes à des outils semi-automatiques pour faciliter la digitalisation de 
différents éléments tels immeubles, routes et rivières. Je ne connais pas 
l'historique des expériences d'utilisation de tels outils. Mais on peut 
remonter en 2011, où on parlait d'un outil de détection de route. Divers 
articles traitent aussi de ce sujet.
https://alastaira.wordpress.com/2011/02/04/automatic-road-detection-using-bing-maps-imagery/https://gis.stackexchange.com/questions/77876/is-there-a-tool-that-performs-automatic-recognition-of-buildings


Facebook a aussi expérimenté des outils de reconnaissance d'image en Thaîlande 
récemment. Selon les plaintes de certains contributeurs, les données ont été 
ajoutées à OSM sans valider suffisamment avec la réalité sur le terrain.
Je pense qu'il serait intéressant pour les contributeurs expérimentés d'avoir 
accès à des outils semi-automatisés facilitant dans JOSM par exemple le tracé 
d'immeubles, routes, cours d'eau, etc. Pour un cours d'eau par exemple, je 
déplace le curseur de la souris, et les contours et le centre de la rivière 
sont tracés automatiquement. Ou encore le contour d'un lac est tracé.
 
Pierre 
 

Le lundi 29 janvier 2018 15:15:59 HNE, john whelan  
a écrit :  
 
 ·    NRCan is working on a methodology to extract building  footprints, 
including topographic elevation and height attributes, from LiDAR


Traditionally OSM has not been happy with this sort of thing.  The accuracy can 
be poor.

We probably need to think about this since the BC2020i project had this 
mentioned and Stats Can has given it a mention also.  I'm not promoting it nor 
saying its bad but it will almost certainly be raised shortly.

First if an import was done using this data who would be the local group to 
approve it?  I suspect because it covers the entire country it would be the 
talk-ca group.  The date would come through the TB portal so licensing is not 
an issue.  Or it could be split into regions with regional local groups making 
decisions.

The other very big question is to do with data quality.  So far nothing that is 
machine learnt from imagery has consistently met the expectations of 
OpenStreetMap.

Note to Pierre I'm not sure if you are on the talk-ca mailing list but any 
feedback you might have on the data quality side would be welcome and will be 
shared amongst the group.

Thoughts?

Thanks John
___
Talk-ca mailing list
Talk-ca@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-ca
  ___
Talk-ca mailing list
Talk-ca@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-ca

Re: [Talk-ca] using image recognition to create building foot prints.

[Stewart C. Russell]

On 2018-01-29 03:15 PM, john whelan wrote:
> ·*NRCan is working on a methodology to extract building 
> footprints, including topographic elevation and height attributes,
> from LiDAR
> 
> * Traditionally OSM has not been happy with this sort of thing.
> The accuracy can be poor.

If you want to take a look at this kind of data, Toronto's 3D massing
data set is derived from LIDAR. It's what we have for building outlines,
and it's not bad at all:
https://www.toronto.ca/city-government/data-research-maps/open-data/open-data-catalogue/locations-and-mapping/#db07630f-252d-f7ae-2dff-8d0b38ec6576

cheers,
 Stewart

___
Talk-ca mailing list
Talk-ca@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-ca

Re: [Talk-ca] using image recognition to create building foot prints.

[Adam Martin]

Can't speak for the rest of the country as such, but I do know that the
imagery quality in Newfoundland is rather poor. Recognizing building shapes
would technically work, but the result would be poorly aligned boxes, not
accurate house shapes. We also do not have a local working group as far as
I am aware so any importation would need to be brought to the whole list or
such a group created.

On Jan 29, 2018 4:47 PM, "john whelan"  wrote:

> ·
>
>
> *NRCan is working on a methodology to extract building footprints,
> including topographic elevation and height attributes, from LiDAR*
>
>
> *Traditionally OSM has not been happy with this sort of thing.  The
> accuracy can be poor.*
>
>
> *We probably need to think about this since the BC2020i project had this
> mentioned and Stats Can has given it a mention also.  I'm not promoting it
> nor saying its bad but it will almost certainly be raised shortly.*
>
>
> *First if an import was done using this data who would be the local group
> to approve it?  I suspect because it covers the entire country it would be
> the talk-ca group.  The date would come through the TB portal so licensing
> is not an issue.  Or it could be split into regions with regional local
> groups making decisions.*
>
>
> *The other very big question is to do with data quality.  So far nothing
> that is machine learnt from imagery has consistently met the expectations
> of OpenStreetMap.*
>
>
> *Note to Pierre I'm not sure if you are on the talk-ca mailing list but
> any feedback you might have on the data quality side would be welcome and
> will be shared amongst the group.*
>
>
> *Thoughts?*
>
> *Thanks John*
>
> ___
> Talk-ca mailing list
> Talk-ca@openstreetmap.org
> https://lists.openstreetmap.org/listinfo/talk-ca
>
>
___
Talk-ca mailing list
Talk-ca@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-ca

[Talk-ca] using image recognition to create building foot prints.

[john whelan]

 ·


*NRCan is working on a methodology to extract building footprints,
including topographic elevation and height attributes, from LiDAR*


*Traditionally OSM has not been happy with this sort of thing.  The
accuracy can be poor.*


*We probably need to think about this since the BC2020i project had this
mentioned and Stats Can has given it a mention also.  I'm not promoting it
nor saying its bad but it will almost certainly be raised shortly.*


*First if an import was done using this data who would be the local group
to approve it?  I suspect because it covers the entire country it would be
the talk-ca group.  The date would come through the TB portal so licensing
is not an issue.  Or it could be split into regions with regional local
groups making decisions.*


*The other very big question is to do with data quality.  So far nothing
that is machine learnt from imagery has consistently met the expectations
of OpenStreetMap.*


*Note to Pierre I'm not sure if you are on the talk-ca mailing list but any
feedback you might have on the data quality side would be welcome and will
be shared amongst the group.*


*Thoughts?*

*Thanks John*
___
Talk-ca mailing list
Talk-ca@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-ca