[OSM-talk] Maximum segment length?

Scott Walde scott at waldetech.ca
Thu Mar 22 03:16:58 GMT 2007


Lars Aronsson wrote:
> Scott Walde wrote:
>   
>> But, this causes me to ask the question... how long can a 
>> segment be?  Around here, highways can go on in a straight line 
>> for many, many kilometers.  (One problem stretch was about 
>> 50km.)
>>     
>
> This is related to the use of Bezier curves and map projections.  
> When you have a line segment, you say node A and B should be 
> connected.  But how?  If the two nodes are at the same latitude, 
> should the connecting road run at a constant latitude?  

Yes. :-)  (I do see the ambiguity, though.)
> That means 
> it will look straight on a Mercator projector map, but on the real 
> Earth surface it will be a long, smooth curve (unless that 
> latitude is the Equator).

Exactly.  That's how our east/west roads are built here.  Our 
north/south try to follow longitudes, but require correction lines every 
so often.  Of course, if I were mapping all the grid roads, we wouldn't 
be having this conversation, as I'd have a node every 1.6km.  (1 mile)  
(A huge area of my province is covered by a grid of roads spaced on 1 
mile centers.)
>   For reasonably short segments (say 500 
> metres), the difference will be far smaller than the GPS accuracy 
> (5 metres), so it doesn't matter to us.  But how much off can the 
> middle of a 50 km segment be?  

None, as explained above.
> Do you even know how to compute 
> that?

If you mean, how much off a shortest distance connection would be from a 
"follow the latitude" connection, I'm sure I could have a few years 
back, but it's not the type of thing I do daily, so it would probably 
take me a while.  Basically, it's just calculating the arcs made by 
intersecting two planes through a sphere.
>   Or should you stick to shorter segments?
>   

Not for this reason.  To keep the renderers happy, I'm happy to keep my 
segments shorter.  It would be handy, however, to have a function in 
JOSM that would automagically break a long segment into a series of 
appropriately short segments.  It would be simpler, and cleaner than 
manually eyeballing adding nodes to a long segment, or alternatively, 
trying to build a 50km straight line piece by piece.
> We once (April 2006) had (and maybe we still have?) somebody (Jörg 
> Osterman) who checked the weekly dump for long segments (there 
> were plenty in Russia), so that we could go in and add nodes to 
> make segments shorter

That's too bad, really.  Long segments are efficient.  The problem, 
really, is that we're trying to render small parts of the map -- and 
that the software seems to be built by people who live in countries 
smaller than my 50km segment.  ;-)

ttyl
srw





More information about the talk mailing list