<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 TRANSITIONAL//EN">
<HTML>
<HEAD>
<META HTTP-EQUIV="Content-Type" CONTENT="text/html; CHARSET=UTF-8">
<META NAME="GENERATOR" CONTENT="GtkHTML/3.18.3">
</HEAD>
<BODY>
Hi all,<BR>
<BR>
I'm trying to extract a urban road network from Quickbird imagery. <BR>
The first question is: did someone already something similar? is it really possible?<BR>
<BR>
Now the technical questions: I have the RGB and NIR with 2.4 m spatial resolution and PAN with 0.6 m spatial resolution.<BR>
I followed this procedure: <BR>
<BR>
- i.group to make a group with the RGB imagery<BR>
- make classes by digitizing polygons to identify the probably road and no-road pixels<BR>
- i.gensingset to make the statistic and produce the signature file<BR>
- i.smap to classify<BR>
- r.mapcalc to extract the road class<BR>
<BR>
First problem: with this method I can extract the roads (approximately...) but I get many pixel too which are not road. I would need to made a structural classification too. Maybe I should use r.texture but I don't know exactly how.<BR>
<BR>
Second problem: when I get finally a good classification, a simple r.thin and r.to.vect would not be enough because the pixel are not all close and precise on the lines, but are "scattered" along the line, so I don't get a line but many little disordered lines along the road. Is there a way to get a single and straight vector line in this conditions?<BR>
<BR>
regards,<BR>
tommaso<BR>
<BR>
<BR>
</BODY>
</HTML>