Complex intersections
Geplaatst door Circeus op 12 oktober 2008 in het English. Laatst aangepast op 13 oktober 2008.FWIW I've typically dealt with the simpler cases of separated roads crossing each others or single-carriage road by making all ways cross each others in a single point, which is marked with a highway=traffic_signal
tag. To me, anything else is at best misleading, at worst downright incorrect.
ETA: Apparently, some other people seem to agree with my take.
Overleg
Reactie van alv op 13 oktober 2008 om 05:41
1) Do the lanes in the separated directions curve into the intersection? No, everyone drives straight.
2) Do the lanes in the separated directions touch and overlap each other in the intersection? No.
I really urge these should be drawn "as two parallel crossings". Location of the traffic_signals is not yet used for anything and they might as well be added as separate nodes on all incoming ways at the exact point where they are.
Reactie van Circeus op 13 oktober 2008 om 06:27
Technically, the lanes DO cross as all vehicles in all direction move through the same space. Marking up to 4 independent traffic signals borders on insanity (why would we even want to mark four independent traffic poles?). I suspect there is no fully satisfactory method of drawing those. And I have an even more complicated case not far from home where two street that do not face intersect a double carriage at a single intersection, which neither solution is able to deal with meaningfully IMHO (here).
Reactie van alv op 13 oktober 2008 om 08:36
Anyone driving from north to south never touches or crosses the lanes for driving from south to north. There's space in the center of the intersection that is used for driving east-west (and turning), but only because the nondivided road crosses and connects the separated roads. Keep straight roads straight has been in the editing standards as far as the wiki history goes and the recommended practice in all mailing list discussions. If the crossing road didn't exist, there wouldn't be a turn in the separated carriages so there shouldn't be one when the crossing road is added.
And I didn't mean that one should draw each traffic signal pole, but one could. Eventually someone might be adding pedestrian crossings and having the traffic signals drawn as being only after the pedestrian crossing would then seem illogical.
As for the second example I admit that neither is optimal but having both (the non facing) roads drawn straight across the bigger road is IMO closer to reality: driving across the bigger road includes first turning (almost) parallel to the bigger road and only then turning to the other side. At some point when the offset of the non facing roads get smaller it becomes reasonable to connect them directly, i.e. the crossing just south of your example looks like one. A driving direction "drive straight" wouldn't be confusing in that when crossing the bigger road, but the example intersections is more accurately guided as "turn right and immediately left".
Reactie van daveemtb op 13 oktober 2008 om 09:08
I'm with alv (and the rest of the openstreetmap community) on this.
If you consider driving across a dual carriageway at an intersection, you first cross the path of one carriageway, then then other. There is no single point where both go. They do not technically cross. They are linked by a section of road, which can be considered as the intersecting road.The example you showed simply doesn't reflect the road layout or the path that vehicles take. I would argue that the method you propose is both misleading and incorrect, whereas the current widespread practice is neither.
I believe that accuracy of road layout is more important in openstreetmap than the number of traffic signal nodes. I think it is better to put the traffic signal nodes where they are encountered by intersection users (although ideally we need some way of indicating whether they only apply to one direction of flow for this) and leaving the renderer to simplify this at lower zoom levels through collision detection.
Reactie van Circeus op 13 oktober 2008 om 18:18
I actually just found another intersection done with my method, and not by me Although the ways OBVIOUSLY don't cross (WTF? The way they are organized no sane person would analyse as the two parallel ways crossing!) they clearly do share a node at the intersection by the very criterion we use to decide whether two separate ways are appropriate: there is no grade separation at the intersection, ergo the parallel ways share a node at the intersection.
Reactie van RichardB op 13 oktober 2008 om 22:05
Sorry, I'm in agreement with daveemtb and alv here. For instance, see one I mapped the other day.
osm.org/?lat=53.16474&lon=-2.21032&zoom=17&layers=0B00FTF
A dual carriageway remains a dual carriageway - even if the central reservation/divider/median etc. is not continuous for short lengths. You can think of the bits where the central reservation is not continuous as roads crossing both carriageways - this will also facilitate turning using routing software. Obviously it's a judgement call - but if the gap between two sections of central reservation is short - e.g. a normal road width plus extra room to allow all turning traffic to manuoever easily - then it should still be tagged as dual carriageway really.
In my example - the gap is around 25m (~82ft). The width of the road that crosses the dual carriageway is around half that.
Much bigger gap than that then by all means tag as single carriageay - but I would say if you tag it as a single carriageway - then it would have to be long enough that you'd actually put a way there - not just a meeting at a node.
Reactie van djo0012 op 14 oktober 2008 om 03:44
"In my example - the gap is around 25m (~82ft). The width of the road that crosses the dual carriageway is around half that."
you talk about 25m gap but Circeus talk about an half meter gap, it's so short that we almost doesn't see it on imagery (look at the last link from Circeus it's on the same street (or at least same pattern of street).
Reactie van Circeus op 14 oktober 2008 om 03:49
FWIW, of COURSE I would not use that system for intersections like those on Pierre-Bertrand! There are even U-turn lanes (not shown, I think they should be highway=service, not secondary_link)! A major difference is that it frequent for such signals to actually be independent from each others. Which is NOT the case in the intersections on, e.g. Hochelaga.
Reactie van alv op 14 oktober 2008 om 07:23
So far adding the traffic_signals is much less important than having the geometry of the crossing ways sufficiently correct. If the signals are just marked on the various crossing nodes, someone can then later use a relation to tell routers about the synchronizations and such about all the traffic signals in one intersection, once someone (else) has proposed an adequate relation osm.wiki/index.php/Relations/Proposed/Traffic_Lights for such use.