[OSM-talk] Duplicate Buildings

[Pierre Béland]

Using id's from the Frederick quadruplicate list, I imported the osm metadata using overpass. Note thate the negative values in the list represent relations.

The table below shows that the majority of quadruplicate cases implicate only one contributor. Simple Building duplicates from the same contributor are the easy ones to correct.  It is different for the relations or combination of relations/ways where it is sometime necessary to revise the tagging as a contributor misused duplicates to represent various levels of a building.

osm type	nb of 	     quaddup	objects
            contributors      cases
relation	1		 68	  272
relation	2		  9	   36
way		1		838	 3352
way		2		134	  536
way		3		  5	   20

See https://www.openstreetmap.org/relation/13965412 that contains way/677238859 with inner role to 6 relations that represent each level of the building.

There are also strange schemas. For this block, one building is represented with one way (6 nodes) and 3 relations in which this way has role=outer.
  way 1137657546 		building=cabin
  Relation : 15538065		building=yes
    Relation : 15538064		building=yes
    Relation : Horsnæs Fangststation (15538063) place=locality

These are four different relations.
-10326414	-10326413	-10326412	-10326411
They all share these  2 buildings  as outer members.
48002128	505561207


Pierre


------- Original Message -------
Le samedi 11 mars 2023 à 07:35, Frederik Ramm <frederik at remote.org> a écrit :


> Hi,
> 
> I think an automatic fix of the problem is possible, however it would be
> a good idea to try and find out what the root cause of the problem is -
> bad software, bad imports, bad instructions?
> 
> To get an idea of how big the issue is, I did this on a standard
> rendering database:
> 
> create table buildings as (select way,osm_id from planet_osm_polygon
> where building is not null)
> 
> select a.osm_id, b.osm_id into duplicates from buildings a, buildings b
> where a.osm_id < b.osm_id and a.way ~= b.way and st_equals(a.way,b.way);
> 
> This took a few days - probably it could have been done more efficiently
> - and resulted in a list of about 70k buldings world-wide that are exact
> duplicates (geoetry-wise) of other buildings. The list is here:
> 
> http://www.remote.org/frederik/tmp/duplicatebuildings.csv
> 
> Some buildings are in OSM three or four times (contained i nthe above in
> the form of "a is duplicate of b, b is duplicate of c") but I've
> extracted them in extra files:
> http://www.remote.org/frederik/tmp/triplcatebuildings.csv and
> http://www.remote.org/frederik/tmp/quadruplicatebuildings.csv)
> 
> I don't have the time to analyse the situation in more detail at present
> so if anyone wants to take the above lists as a basis for deeper analysis...
> 
> Cheers
> Frederik