[mkgmap-dev] [PATCH v1] Optimize filters for shapes

Sun Jan 13 08:18:15 GMT 2013

> WanMil wrote
>> 2nd change: The RoundCoordsFilter does not care if an area is rounded to
>> a line which happens quite often. The patch tries to remove the line
>> parts.
>
> please check: the patch ignores preserved points, so I think it could break
> routing.

You cannot route over polygons. So I don't expect that removing a point 
from a polygon can break routing?

>
>
>
> WanMil wrote
>> Anyhow I think there is more improvement possible. I have attached two
>> gpx files. Bye the way the patch contains some commented lines that
>> creates the GPX files for each polygon treated by the patch.
>> 11315_16_org.gpx is the input polygon to the RoundCoordsFilter and
>> 11315_16.gpx is the result (using the patch).
>>
>> Some possible improvements:
>> * Straight lines sometimes contain more that one point and it should be
>> easily possible to remove these additional points.
>> * The filtered polygon sometimes still contains lines which are not
>> removed by patch because the overlapping lines do not contain the same
>> points. You can see this in the attached gpx in the upper middle of the
>> right polygon block.
>> * The filtered polygon could be simplified and sometimes split into more
>> separate polygons which are now connected by a line.
>> * Better resolution dependend filtering?
>
> I would assume that straight lines are filtered by DouglasPeuckerFilter ?

Probably not. Otherwise the file size of the img files should have been 
the same.

> (at least now that I've committed the removeShortArcs patch)
> I did already look for other algos, eg. the VW algo
> http://www2.dcs.hull.ac.uk/CISRG/publications/DPs/DP10/DP10.html
> sounds like a possible alternative to DouglasPeucker.
> On the other hand, I assume that the input for the filters will change
> with the intended mp_cut changes, so we may end up solving the same problem
> at two different places.

Yes and no. The mp_cut changes will only affect how polygons are cut 
when they are part of a multipolygon. But there are many polygons that 
are not.
On the other hand I also use the algorithm of the RoundCoordsFilter in 
the mp_cut branch (not yet committed) and therefore observe exactly the 
same problems. So improving the RoundCoordsFilter algo will also solve 
problems in the mp_cut branch.

WanMil

>
> Gerd
>