[Elphel-support] [Elphel-support-Eyesis] Please (stereo) rectify me ...

[support-list-eyesis]

Luc,

We did not use the 2 bottom cameras for panoramas, but they are treated exactly the same and I remember them showing on the overlap maps correctly. Can you be more specific on "unrelated" or even place all images as layers in gimp using those two values and send the composite image?

As for processing image pairs - there is some code in ImageJ plugin (I made couple years ago) that can map 2 images on a common plane (common defined as average of the two normal vectors) and oriented so disparity is strictly horizontal.

Andrey

---- On Mon, 25 Nov 2013 16:38:24 -0800 Luc Deschenaux<luc at miprosoft.com> wrote ---- 


  Hello Andrey,
 
  For channels 24 and 25, the XPosition and YPosition properties seems unrelated to the upper channels panorama coordinate system.
 
 Then to estimate azimuth for pixels in channel 24, i tried to use XPosition and YPosition from channel 10, then to add channel[24].azimuth - channel[10].azimuth, with more or less precision.
 
 I believe that because of the horizontal rotation between top and bottom cameras, for near objects there's a small horizontal disparity as well as vertical disparity.
 
 What do you have in mind to achieve proper rectification of the two stereo pairs in order to align epipolar lines with the baseline and to compute distances at short range from matched pixel coordinates ?
 
 After reading "Stereo rectification of calibrated image pairs based on geometric transformation
 " (http://www.mecs-press.org/ijmecs/ijmecs-v3-n4/IJMECS-V3-N4-3.pdf) I believe if it's worth trying this method which boils down to compute two projection planes coplanar and parallel to the baseline and reprojecting pixels on them (and I do try to implement it). What do you think about it ?
 
 In the same vein, for a matched pixel in two equirectangular images, after estimating relative azimuth and elevation with respect to the equirectangular image center, i believe it is possible to map the given vectors coordinates relatively to the baseline in order to estimate depth but i did not succeed yet...
 
 ...so i wonder: Is there is a scale factor or something that makes the focal length (in pixels units) no longer equal to the focal_length_mm / sensor_pixel_size_mm when working in the equirectangular image coordinate system ?..
 
 Thanks in advance for your answers !
 
 Luc
 
 PS: In elphel-imagej/PixelMapping.java I see there's a "Create Plane Map" procedure generating .plane-tiff and .intermap-tiff files. I wonder what are those .plane-tiff and .intermap-tiff files, and, if they are really useful for what i'm trying to do, what are the required steps to generate them ?
 -- Luc Deschenaux t: +41 788 266 366 v: +41 225 085 085 @ netvoip.ch _______________________________________________ 
Support-list-eyesis mailing list 
Support-list-eyesis at support.elphel.com 
http://support.elphel.com/mailman/listinfo/support-list-eyesis_support.elphel.com 



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://support.elphel.com/pipermail/support-list_support.elphel.com/attachments/20131125/c4f50925/attachment-0002.html>