[Talk-GB] building shapes from OS Street View

Ian Spencer ianmspencer at gmail.com
Mon May 24 15:29:01 BST 2010


Ed Avis wrote on 24/05/2010 15:05:
> It appears the choice for buildings is a fight between OS maps, which are likely
> derived from high-resolution aerial photos but have been simplified, and the
> somewhat lower-resolution photo images available to OSM.  It is not really
> possible to survey building shapes on the ground since GPS isn't precise enough,
> even if we had enough mappers to walk round the outside of every building.
>
> In London the Yahoo photos are adequate for tracing streets, but trying to get
> buildings from them is a bit ropey.  Hence my feeling that the OS data is of
> higher quality, even though it's clearly a simplification of the real shape.
> But in other, less crowded parts of the country a semi-automated trace from
> aerial photos (using the Fuzzer JOSM plugin you mention) might be a better way.
>
> (Indeed, it might be fun to trace the whole country using both methods and then
> highlight differences between the two.)
>
> I still believe we could do *something* to get reasonably good building shapes
> into the map without waiting for the man-years needed to hand-survey them all.
> Even in an area which is not any mapper's particular 'territory'.  The workflow
> might be to view the aerial photograph with the OS-traced building shapes on top,
> and then either press a button to import the whole area or select individual
> shapes.
>
>    
Taking the lead from StreetView in my area, the shapes around are better 
than nothing and are good for making sense of the map. With appropriate 
tagging, which is trivial with a bulk import, then it becomes easy 
enough to have a to-do list. Areas of little interest to people will be 
populated adequately, and areas where people are interested, they should 
know from the source (perhaps even an automated note) that they have 
carte-blanche to improve it.

A Wiki-Map does have the danger of trying to be all things to all men, 
and it is interesting to see from a single source that the cyclists can 
produce an appropriate map different from the general. If people are 
really uncomfortable with the impurity of the data (in the sense of 
reliability not licence), then the right approach to me is to have a 
specific tag that is internationally agreed to be unverified/suspect and 
then map producers can filter or represent that data as they deem 
appropriate, without the map editors needing to prejudge an appropriate 
solution. It would also be useful to then make PotLatch et al sensitive 
to the tag. (BTW, it may be that this already exists, but there does not 
seem to be anything from my cursory look).

That is not just for this extracted data, there are times I have looked 
at an area I know well and can't believe it does not mark a path that I 
know well. I would rather put the path in temporarily with a warning 
that it needs to be verified for exact alignment, and gets on to a to do 
list, than not mark it at all. I'll walk it eventually, but on a map it 
might get noted and verified by someone else.

Spenny




More information about the Talk-GB mailing list