[Talk-GB] Traffic lights

[Dave Stubbs]

On Mon, Sep 15, 2008 at 6:27 PM, Richard Bullock <rb357 at cantab.net> wrote:
>>
>> I'm of the school of opinion that places bus stops to the left/right
>> of the way but for traffic lights I do exactly what you do.
>>
>> - It is easy to enter and render
>>
>> - It is easy to extrapolate backwards, i.e. one
>> highway=traffic_signals on the intersecting node equals one light on
>> each way coming into the node (even taking into account they may be
>> one-way).  There is no practical need to know which side of the road
>> the lights are(?). I'd be happy to mark lights individually if I find
>> a special case, but I haven't yet.
>>
>> - It is adequate for routing (?), i.e. if a way has a traffic_signals
>> node on it, add a penalty.
>>
>> - For routing, if there is a left filter without a light, then I draw
>> a small connecting way short-cutting the lights node.
>
> That seems reasonable to me. I normally place a single
> <highway=traffic_signals> on the node of intersection.
>
> I have found some special cases - e.g. where a road narrows to cross a
> bridge or go under a tunnel - and traffic flow is controlled by having a
> traffic light at each end - e.g. this one either side of a narrow railway
> bridge near to Chelford in Cheshire.
> http://www.openstreetmap.org/?lat=53.2774&lon=-2.26827&zoom=17&layers=0B00FTF
>
> I'd be interested to know if anyone would tag that differently - given that
> the traffic signals point in one direction only.
>

The other special (actually not so special) case is where you're
trying to indicate a pedestrian/cycle crossing. Particularly with
toucans, not all roads feeding an intersection will be crossings so
you want to mark the actual location of the crossing -- the problem
then is you have the toucan crossing before the traffic lights on the
way, which doesn't make a lot of sense. Of course if you mark all the
crossings in with traffic lights (non directional as a node) then you
have two lots of traffic lights for every way across the intersection,
which also doesn't make much sense. I'm guessing this is just one of
those cases where routers/renderers are going to have to be vaguely
intelligent about proximity -- at least all the nodes are referenced
by the ways already, so no need to go looking for them.

Dave