[OSRM-talk] Worldwide coverage of Osmose-QA

Daniel Hofmann hofmann at mapbox.com
Tue Aug 2 09:26:38 UTC 2016

On Mon, Aug 1, 2016 at 8:49 PM, Frédéric Rodrigo <fred.rodrigo at gmail.com>

> Le 31/07/2016 à 21:34, Daniel Hofmann a écrit :
>> > We are still open to contribution on analyser code or suggestion, on
>> somewhat poor documentation on the wiki, on translation or whatever else.
>> With the focus on guidance and lanes during the last weeks and months,
>> here are some cases I can think of.
> We already have some checks on lanes
> https://github.com/osm-fr/osmose-backend/blob/master/plugins/Highway_Lanes.py
> http://osmose.openstreetmap.fr/en/map/#layer=Mapnik&zoom=12&lat=38.8967&lon=-77.0161
> But it's a complicated subject.
> (An issues too https://github.com/osm-fr/osmose-backend/issues/115 )

Great to see some lane checks! We know lanes are difficult that's why QA
for them is all the more important!

- Lanes for roundabouts are a mess (ref.
>> https://github.com/Project-OSRM/osrm-backend/pull/2693) - we looked into
>> this and came to the conclusion to not support lanes inside roundabouts or
>> do lane anticipation for lanes in roundabouts at this point in time. Maybe
>> we can come up with some automated checks here.
> As you experienced with the subject, can you propose a check ?

I think the most simple check would be to detect roundabouts that have
some  kind of lane tags on them. And then making sure the ways directly
leading into a roundabout and the ways right before the exit have proper
lane tags, or lane tags at all.
This would already make lane handling for roundabouts in the routing
context much more practical.

> - Multiple lane markers on the same lane as in left;left|right (ref.
>> https://github.com/Project-OSRM/osrm-backend/pull/2697) - you probably
>> have to check this in detail as in: are there situations and/or countries
>> where it makes sense to have multiple times _the same_ lane marker on a
>> single lane.
> What means "left;left|right" ? And what can be wrong with this ?

left;left|right means you have two lanes, the leftmost lane has two arrows
to the left (does this make sense at all? always mis-tagging?), the
rightmost lane has one arrow to the right.
Example for this tagging scheme:

- Advanced lane features depend on lane information right before the turn
>> (e.g. Lane Anticipation ref.
>> https://twitter.com/ericg/status/757887419530248192), maybe an automated
>> check could test if there is lane information on/after e.g. motorway links
>> but not directly before them as per the lane tagging scheme.
> Do you talk about the transit:lanes proposed feature ?

Interesting, I haven't heard about this before, thanks for mentioning it!
I implemented Lane Anticipation already in OSRM without this tag, based on
turn lanes and route context OSRM has at the intersections.
For this to work, the route steps involved have to have lane information
associated with them.

- Lane access tag values not matching turn lane tag values.
> We do not already make inter lanes tag check yet, but make sense.
> - Impossible or strange lane configurations such as the left most lane
>> turning to the right crossing straight/none lanes (in right-sided driving
>> countries).
> This one, at least, look easy.
> - Roundabout mis-tagging (ref.
>> https://github.com/Project-OSRM/osrm-backend/issues/2716): max. radius /
>> outlier nodes in roundabout way.
> You suggest to detect gap between a node and the local shape of the
> roundabout ?
> After micro-mapping, this is micro-fixing !

Haha, I think a good first check would be to detect outlier such as this

I highly appreciate what you are doing, thanks for taking the time to talk
to us! :)

> Frédéric.
> _______________________________________________
> OSRM-talk mailing list
> OSRM-talk at openstreetmap.org
> https://lists.openstreetmap.org/listinfo/osrm-talk
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/osrm-talk/attachments/20160802/1c137ad9/attachment.html>

More information about the OSRM-talk mailing list