[Openstreetmap] our little meetup today

[Jo Walsh]

this afternoon steve, matt, saul and i met for drinks and talked
openstreetmap. we talked about data interfaces and about possible
intending publicity and how that image might jar with our world image.

- Nick's going to pour the free-map.uk data into openstreetmap via the
  XML-RPC interface

- I'll start using that interface for London/Bristol data that we get.
  (The easton wireless people are collecting tracks to map their
  network, and matt has a good set of tracks for the other side of town.

(steve raised the point, "why did imdb explode, why did wikipedia
explode when parallel projects alongside them didnt? because of
database centralisation and 'completeness'". from my semweb-biased
perspective i don't think about where the data is, perhaps not enough)

http://www.openstreetmap.org/wiki/index.php/XmlRpcInterface
i am unsure about what to do with a point's trackid. manage that
myself and supply an arbitrary value?

- We talked about road ontologies a fair bit:
  - future extensibility is crucial
  - readability is nice
  
  I think matt's ideas about annotating abstract street
  representations (segment collections) with key-value pairs works and
  corresponds pretty well to the RDF model i had in mind. 
  
  the other week i did a simple hack to infer street continuity by
  comparing the angles between touching segments at each end of a
  segment.
  http://www.mappinghacks.com/projects/mapscript/02_roadseg-scrot.jpg

  Matt talked of graph theory in the context of a node-line-shape
  model. now i reflect that i don't have an overt internal
  representation of nodes (junctions), only implictly through geometry
  queries on segment contiguity. 

  My RDF models have had a concept of a Junction node explicitly,
  to which 2+ segments connect. TIGER/Line doesn't treat of junctions,
  again it is a query matter to see if a start or end left or right
  point corresponds to a given point.

- We argued over data purchase vs hardware purchase some more

  My feeling on this is still:
  - somehow we will acquire enough money to do justice to both
  - ikonos can be acted on 'now' and a community site built around
  - we really need to test sample the GA approach before buying

- The 'shortest path to shiny thing' idea surfaces. I am immensely
  keen to see 'shiny thing' at least in glimpse, on the 14th. 
 
Re. GPS trace gathering and line segmentation. Matt mentioned that
line segments shouldn't intersect/overlap, but always break and join
at a junction. in graph terms this makes sense to me.

I found when experimenting with *manually drawing* over gps tracks
(instead of using douglas-peucker etc to estimate better lines;
probably not really a sustainable idea but i was curious)   
that i captured a lot of segments which in RL have many junctions and
should be broken up, but are only known in my GPS observations of the
world to be a contiuous stretch. This 'incompleteness' means that
whenever we get a new cross-secting line, there has to be a post-process
that finds line intersections, breaks both lines and reapplies the
metadata properties to all the new segments, replacing that bit of the
original line. I suppose if there is a 'street' abstraction with a
'stack' we can insert things into the middle of, then this becomes
relatively easy...

looking forward to the 14th and the opportunity to strongly articulate
and implement common goals...


-jo