<html>
<head>
<style><!--
.hmmessage P
{
margin:0px;
padding:0px
}
body.hmmessage
{
font-size: 12pt;
font-family:Calibri
}
--></style></head>
<body class='hmmessage'><div dir='ltr'>Hi all,<br><br>up to now I've implemented <br>a) the algo to calculate the convex hull (using the QuickHull algo) for a set of coordinates <br>b) an algo to identify elements with addr:place tag which are probably referencing the same<br>place. I use a combination of addr:place,addr:city,addr:postcode and the mkgmap:admin_levels8 to 10<br>This makes it less likely that a wrongly tagged object is blowing up the area for a hamlet,<br>but I still see hamlets with a dense cluster of nodes and a few ones rather far away<br>which might be copy+paste errors in the OSM data.<br>c) an algo to the find roads without mkgmap:street which intersect with one or more of the computed convex hulls.<br><br>This is all work in progress and not yet usable.<br><br>Open problems:<br>1) The convex hulls may be overlapping. I assume I have to subtract the overlaps (from all areas)<br>2) Some OSM ways connect different hamlets. I think I have to split them somewhere outside of any of the convex hulls.<br>3) The convex hull might be (nearly) a straight line if only a few houses along the same side of the road are <br>tagged with addr:place. I guess I have to make sure that I'll find at least one road that is close to them.<br>4) I am not sure if I should name all unnamed routable ways with the name of the hamlet. This could include<br>tracks and steps etc. . I think I should check a tag like mkgmap:ignore-for-housenumbers=1<br>to exclude roads from this algo.<br><br>Any better ideas ?<br><br>Ciao,<br>Gerd<br><br><br> </div></body>
</html>