documentation and possible SPW PICC server tuning

[a-pirard]

Using Area Selector with Imagery>Preferences>BE SPW PICC is simply gorgeous !!!
I feel so sad to read people write that they can' find how to use AS that I feel like writing a small doc.
But I would also like you to read a user experience report with that SPW server to see if there are VERY SIMPLE changes that could improve that case.
Could we swap this?

[r00tat]

Thanks for your feedback!

I tried to update the Wiki with a little more information on how it works. I have also added a Configuration wiki page.

[r00tat]

if you have any suggestions, that would improve the usability or the function of areaselector, please let us know!

[a-pirard]

On 2015-10-25 16:58, Paul wrote :

if you have any suggestions, that would improve the usability or the
function of areaselector, please let us know!
Really glad you replied.
Sorry I (user clicking in JOSM plugins page) didn't find your doc updates.

Regarding improvements, the best is for you to make experiments (in
South Belgium) with the following JOSM overlays and to tell your
feelings about possible improvements. Read below.

PICC0:
wms:http://geoservices.wallonie.be/arcgis/services/TOPOGRAPHIE/PICC/MapServer/WmsServer?&SERVICE=WMS&VERSION=1.1.1&FORMAT=image/png8&TRANSPARENT=TRUE&REQUEST=GetMap&STYLES=&SRS={proj}&WIDTH={width}&HEIGHT={height}&BBOX={bbox}&LAYERS=0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,48,51,52,53,54,55

wms:http://geoservices.wallonie.be/arcgis/services/TOPOGRAPHIE/PICC/MapServer/WmsServer?&SERVICE=WMS&VERSION=1.1.1&FORMAT=image/png8&TRANSPARENT=TRUE&REQUEST=GetMap&STYLES=&SRS=%7Bproj%7D&WIDTH=%7Bwidth%7D&HEIGHT=%7Bheight%7D&BBOX=%7Bbbox%7D&LAYERS=0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,48,51,52,53,54,55BASE:
wms[19]:http://geoservices.wallonie.be/arcgis/rest/services/DONNEES_BASE/FOND_PLAN_CARTE_2012_v1_RW/MapServer/export?&f=image&format=png8&dpi=90&transparent=True&SRS=EPSG:3857&bboxSR=3857&imageSR=3857&bbox={bbox}&size={width},{height}

I have used areaselector with PICC0 rather much and the major problem is
extraneous nodes. I remove them (and I sometimes improve precision)
with Improve Way Acuracy Mode, but that makes me sometimes feel like it
would be faster to draw the ways "manually". Extraneous nodes are
mainly caused by the ragged shape of the lines, by the house numbers
overlapping lines, by other lines crossing the buildings and sometime
the middle wall common to two buildings is duplicated.

BASE, on the other hand, is much more like like basemap.at. But it is
coarser (less precise). I did not use it much because there is a
(color?) problem detecting middle walls (makes one house instead or two
or more).
I'm starting a mapping session and I'll try more of BASE.

If PICC0 was really too difficult to improve and BASE was easy, I might
try to ask if there is a finer version of it.

[r00tat]

Could you give me a bbox or coordinates, to try it out in belgium?

[r00tat]

I tested a little bit at 49.964546774722876, 5.447027359130042. PICC0 has a better quality and will likely produce better results. At higher zoom levels, the lines start to look jagged.

With all default settings it will lead to a building which is not exactly like it should, but not to bad.

In my experience the buildings mapped by such imagery like PICC0 always have to be checked with aerial imagery. In some cases you find out that the buildings changed or looks different.

I'd like to increase the accuracy of the plugin, but I'm not sure any more how. I tested a lot, believe me reaaaallly a lot, but did not get to a real solution. In the end I'm also using the simple Canny Edge detector from BoofCV to detect the polygons as the default algorithm and my custom algorithm as a backup.

[a-pirard]

On 2015-11-04 23:18, Paul wrote :


  I tested a little bit at 49.964546774722876, 5.447027359130042.
    PICC0 has a better quality and will likely produce better
    results. At higher zoom levels, the lines start to look jagged.


  With all default settings it will lead to a building which is
    not exactly like it should, but not to bad.

This is what I call a good result.  It could have been improved with
the 90° enforcing, see below.
In my experience the buildings mapped by such imagery
  like PICC0 always have to be checked with aerial imagery. In some
  cases you find out that the buildings changed or looks different.

Mind you, these PICC pictures are made out of this
  aerial imagery plus administrative data. They are simply
fantastic. If they look a bit different, it's because the aerial
pictures are corrected. Aerial pictures are shot from an angle and
hence the vertical lines are slanted. You see the house walls when
you would not see them if the pictures were taken from the top (like
satellites).  What PICC shows is the walls at ground level.
Mappers who use the aerial photos and who are not aware of this map
the roofs and cause an error up to 5m for high buildings.
The PICC map is correct at 0.02 m.

I  have mapped the houses along this road from here  to here and
I have left the errors in for you to see.
But some errors do not show there because most houses are detached.
So, I added a few more here Cour Dogné.
Don't forget to tell me when you've finished looking at that.

Rue de la Croix André:
started odd numbers at 6m/35° auto
changed at 47: 4.5/24 bootfcv
57, 59, 61  unable to find a polygon
changed at 2: 4.5/24 custom
17, 65, 69: JOSM: crossing buildings

I suppose a per house report of the defects wouldn't be useful to
you, you can see them by yourself.

The most frequent errors I see are:
Errors that could be fixed by enforcing a right angle:
- two nodes instead of one at a house corner (like a hammered
corner)
- if the wall is like this ___|¯¯¯, the vertical bar is inclined
(angles > 90°)
Sometimes, some wall is unexplainably too much inside.
If a wall goes across a house number (roundel), a zigzagging way can
result.
If some line crosses the house, the way may follow it (crossing
buildings above).
Where two houses are attached, indescribable entangling may occur.

I may provide you with examples if you ask.
I may extend this list as I watch my doings)
I'd like to increase the accuracy of the plugin, but
  I'm not sure any more how. I tested a lot, believe me reaaaallly a
  lot, but did not get to a real solution. In the end I'm also using
  the simple Canny Edge detector from BoofCV to detect the polygons
  as the default algorithm and my custom algorithm as a backup.
Increasing the accuracy would be welcome, but that's not the most
important problem I meet.
The problem is that AS fails to detect the general structure
(outline) and does bogus drawing.
I've had a discussion with my always enthusiastic friend cc: Alain
and we came to this conclusion.
AS is able to determine the lines of a building with a fair
precision.
But it sometimes mixes them up because it has no notion of how the
building is structured.
The human being sees that very clearly but is awkward about
precision.
So, he could click from corner to corner and AS would snap his
clicks to the nearest line crossing.
The user would try the present, one-click method first.
If it succeeds (many cases), all done, next.
If it fails, undo everything in one click and use the method I
describe.
All in all, it will be a very substantial saving of clicks, because
correcting what AS did when it goes astray is very tedious, tricky
and time consuming.

Well, Paul, I hope this will help you and I really must go to bed.

[a-pirard]

Paul,

  Instead of leaving AS-to-improve examples on OSM,
  I suggest that I copy them to a .OSM file and upload it.
  The name tag will contain a descriptive title and the note tag may
  contain an explanation.
  You are able to experiment with that.

  Do you appreciate that method?

  But please make an update using     pixel-tolerance =
  meter-tolerance ÷ zoom 
  so that the tolerance does not change when zoom changes, 
  Without that, I am unsure or what I am doing.  thanks thanks
  thanks.

  I try attaching, uploading .OSM, else an OSM.TXT file.

  Contained:
  Errors that could be fixed by enforcing a right angle:
  - flat corner: two nodes instead of one at a house corner (like a
  hammered corner)

  - entangled 0: house 5B ignores separation and reuses 4 and 4A
  nodes, but strangely

  To come:
  - if the wall is like this ___|¯¯¯, the vertical bar is inclined
  (angles > 90°)
  Sometimes, some wall is unexplainably too much inside.
  If a wall goes across a house number (roundel), a zigzagging way
  can result.
  If some line crosses the house, the way may follow it (crossing
  buildings above).
  Where two houses are attached, indescribable entangling may occur.
  Increasing the accuracy would be welcome, but that's not the most
  important problem I meet.


Cheers





      André.

[a-pirard]

AreaSelector improvements (copy).osm.txt