|
Posted by smartpointer on March 3, 2010, 3:39 pm
Please log in for more thread options
Hello,
I was wondering if somebody could help me, I've been tussling with
this problem for a pretty long time.
I have a stereopair of a simple shapes in an artificial scene created
using blender 3d tool. I'm creating a disparity map using SSD
algorithm and using opencv Canny filter to get characteristic points.
From these disparities I'm gathering I'm trying to recreate structure
of the shape in 3d.
Here's a screen grab of my gui with a sample cube and the 3d
reconstruction based on
disparity values:
http://img175.imageshack.us/img175/116/sampley.jpg
As you can see instead of smooth lines there are these bundles of
pixels with same
disparity/distance values ( I get distance values using focal length
and baseline formula ) which is pretty obvious because there's a
limited number of disparities assigned to a big number of pixels. I
was wondering if there's any way in reconstructing the nearly exact
shape or at least some kind of an outline without going into line
recognition or pixel tracing algorithms. I heard about using the
planar transform but couldn't find anymore information about how it is
supposed to be used. That link is still pretty useful, neat summary of
the scanline based disparity search algorithms.
Thanks.
frd
| 
XML Sitemap