[Talk-GB] TfL Cycling Infrastructure Database - conflation

[Richard Fairhurst]

Martin Lucas-Smith - CycleStreets wrote:
> Richard will be doing the bulk of the scripting work, and is working 
> on converting each of the sections of data. This will naturally be 
> published on Github openly, as will the outputted data. This is 
> reasonably complex work given the number of attributes and the 
> data extent. We are keen to ensure the OSM community is 
> able to scrutinise the conversion easily and have input. Richard 
> will post to this list about the work, as it proceeds.

A quick update on this one:

Hoo boy, there's a lot of data!

The good news is that it's consistently high quality, and lots of it isn't
in OSM already. Our cycle parking and speed bump coverage, for example, will
be massively enhanced by this. In a few of the datasets, particularly
(traffic-free) cycleways, we already have most of the information but the
TfL data has identified missing items - for example, a decent shared-use
path (almost a mile long) beside Snakey Lane in Feltham, shared-use paths
beside the A1 in Mill Hill, and so on.

The line geometries (cycleways and footways) will require a fair degree of
manual work to get into OSM, obviously to ensure connectivity but also for
sanity checking. The point data (cycle parking, traffic calming) etc.
varies, and I've been classifying the output as "easy new data" or "needs
further review" accordingly.

I'll push some code and output to Github in the next few days so people can
have a play. The general approach is that the TfL data is compared against
an OSM PostGIS database, which means we have the full power of PostGIS's
spatial analysis to help match features. Don't underestimate how complex the
matching is: I've been working for 8+ days on it (lockdown's not such a bad
thing...) and it's not finished yet. Currently I'm outputting GeoJSON for
easy visualisation, but depending on the conflation tools eventually used,
there'll almost certainly be .osm files too. 

There is an elephant in the room, and that's the (on-road, painted) cycle
lane data. Once again, this is really high quality data. It's all new tags
on existing ways (because we have all the roads mapped), but because it'll
mean splitting the ways to get the extents right, it'll be a challenge for
conflation. I'm currently thinking through the best approach for this, but
again, I think it'll ultimately involve classifying the data into levels of
confidence: "this is already in OSM", "this can be added easily", "this will
need further review", etc. etc.

As Martin set out, TfL have dedicated some time to training up their staff
and working on the data. That'll work well for cycle parking etc., but the
in-depth work on cycle lane attributes is almost certainly going to be best
done by experienced OSM mappers. Once I've got the first set of output up on
Github it would be great to take a steer from interested people as to what
they'd find most helpful.

cheers
Richard



--
Sent from: http://gis.19327.n8.nabble.com/Great-Britain-f5372682.html