2d -> 3d reconstruction

Hi,

I am developing a 3D reconstruction system using 2 cameras in a stereo
rig configuration.  Calibration is achieved using the direct show
calibration filter (CvCalibFilter) included as part of the OpenCV
library.  Stereo matching across the input images is achieved on a
pixel by pixel basis using Gabor Jets as the correlation measure. 
Matching is   done on the non-rectified input images.  Thus input into
the reconstruction stage includes full intrinsic and extrinsic camera
parameters and a set of (unrectified) matching image coordinates.  I
have implemented the reconstruction algorithms as described by several
books on the subject, however, I am not getting reliable results.  If
anyone here has worked in this field perhaps they could point out
where I may be wrong since the code produces poor/totally incorrect 3D
reconstructions even given correct calibration parameters and point
matches...

I have included the source code for the construction of my projection
matrices and the code that solves the resultant linear system to
produce the x,y,z coordinates from the x,y pairs.  In simple cases
(such as reconstructing synthetic stereo pairs, where cameras are
perfectly parallel) apparently correct results are produced however in
more complex systems with arbitrary camera configurations the results
are clearly incorrect.  I can provide my calibration and point match
data should anyone here find the time to help me out and obviously
anyone can use my reconstruction code ;-)

Current reconstruction source code here:
http://cs.nott.ac.uk/~djb/Reconstruction.cpp

Cheers for any help,

Daniel Bardsley 





Change settings: http://www.yahoogroups.com/mygroups, select
   Get Emails (get all posts)
   Daily Digest (one summary email per day)
   Read on the web (read posts on the web only)Or Unsubscribe by mailing 
OpenCV-unsubscribe@xxxxxxxxxxxxxxx