Hi,
Jon Stockill schrieb:
[SNIP]
Here's a couple of pics, the first is looking west over the gherkin, and
the second is looking out over regents park. Generation time was over an
hour for that tile on a 1GHz athlon (the resource limits in
fgfs-construct needed a significant increase). Memory usage was ok at
around 140MB.
http://flightgear.stockill.org.uk/scenery/osmroads1.jpg
http://flightgear.stockill.org.uk/scenery/osmroads2.jpg
This looks quite detailed. I'm not that familiar with the London area so
would you say that there is a considerable amount of smaller streets
missing in there?
Even with not so much detailed streets - i.e., leaving all the
back-streets out and having only freeways and mainstreets processed -
the generation of scenery takes substantially longer than for the
standard data. I have no numbers on this, but having to raise the
resource limits in fgfs-construct considerably is quite a good indication.
One other problem that'll probably become more serious with more
detailed street-data is that the comparatively small streets not only
produce lots of small triangles on the streets themselves but also raise
the number of triangles on neighbouring polygons by splitting them.
(Currently fgsd crashes on loading some of the more "full" tiles -
possibly because of the sheer number of triangles; I will investigate
further on that when I get the time)
I currently do not have any solution for that and I know that there have
been discussions before, but perhaps having more detailed scenery now
may be a good reason for the experts here to reconsider this topic.
It shows up a few limitations in the source data. The road segments are
currently all represented seperately and not tied into a road object,
this results in lots of short roads, and visible breaks on the outside
of corners. This will improve as the open streetmap system matures (it's
still very early days).
This may be solvable by automatically snapping endpoints and recombining
shorter segments into longer segments.
Anyway recombination will only partially solve the problem, as on an
intersection only pairs of participating line-segments can be joined.
Perhaps we need a better way of making polygons from lines, much like
the v.buffer operation of GRASS
(http://grass.itc.it/grass61/manuals/html61_user/v.buffer.html).
Regards,
Ralf
_______________________________________________
Flightgear-devel mailing list
[email protected]
http://mail.flightgear.org/mailman/listinfo/flightgear-devel
2f585eeea02e2c79d7b1d8c4963bae2d
