[OSM-talk] Why we square buildings (WAS: iD invents nosquare=yes for buildings which should not be squared)

Pierre Béland pierzenh at yahoo.fr
Sat May 11 20:27:37 UTC 2019

Am 11/05/2019 um 21.09 schrieb Simon Poole:
> Just a general remark on the technical issue that sparked of this
> discussion:  squaring buildings is not primarily about improving data
> quality. Non-square buildings are simply visually annoying when
> rendered, so much that I support squaring them "as a rule" too when it
> can reasonably be assumed that there are 90° angles in the actual
> building outline. But I'm not kidding myself that it improves "quality".
> If we wanted to define quality of building outlines in OSM we would
> probably be talking about deviations from the buildings footprint area,
> average deviations from the outline and so on, any of these could be
> very small even without squaring. Actually, squaring might impact them
> negatively, particularly when the outline is rough, but as said: square
> buildings are just so much easier on the eyes :-).

See some annoying deviations from the building footprints we would prefer to no observe on the map
High ratios of Unsquarred buildings is often an indication of imprecise mapping with often significative deviations from the outlines. But, yes this is more then worry about squarred buidlings. We should also worry about the general problem of deviations from the footprint.  Dark and unaligned Imagery, various images with different offsets and inexperience in correcting the offsets also contribute to bad mapping. 

An angle deviation of 1-2 degree is surely acceptable. But analysis of mapping shows in some areas very high ratio of unsquarred buildings or irrregular Round buildings with important deviations. We see the same with Building Imports projects. We cannot apply a strict Yes or No rule like for Topology errors. But how much shoud we deviate ?  Geography or Pop Arts ? 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/talk/attachments/20190511/4193eb24/attachment.html>

More information about the talk mailing list