$Id: changes.txt,v 2.2 2005/05/24 09:15:31 svoboda Exp $

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

List of the most significant changes since the last public release
from 30th July 2003

- The RANSAC validation was rewritten as m-functions. Hence, no
  re-compilation of the c-codes is necessary. The user can switch
  between the mex- and m- version of the RANSAC by specifying the path
  in gocal.m addpath ../RansacM; % ./Ransac for mex functions The
  m-version can be slow for highly noisy data.

  The m-version of the RANSAC uses the 8-point algorithm not the
  7-point algorithm as the c-version. It is then less optimal. I was
  just too lazy to rewrite the more complicated 7-point algorithm. I
  did many experiments, the 8-point algorithm did well in all cases. 

- gocal.m ... code cleaned, output better formatted (new function
  dispcamstats)

- local alignments ... code cleaned, redundant I/O parameters
  (possible source of errors) removed

- getpoint.m ... code cleaned, made a bit faster. Made compatible with
  the newest image processing toolbox, ver>5

- showpoints.m ... it possible to switch between single images and
  image mosaics. The image mosaics may sometimes speed-up the
  debugging process when something goes wrong with the point
  detection.  

- The standard Matlab function quiver was copied to the BlueCCal
  distribution since it has problems in the newest Matlab ver>7.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

List of changes against to the last public release from 18th July 2003:

Fixed Bugs in Code:

- radial distorion ... proper handling of images with resolution
  different than 640x480


Functionality:

- support for non-color images added. You may either specify
  config.imgs.LEDcolor='intensity' and the color images will be
  converted to a grayscale or the GETPOINT detects that the image is
  only grayscale and it will be handled as such

- main cycle in GOCAL. Some upgrade to make it more robust. I am
  testing the upgrades on all datasets I have. It should be backward
  compatible. Typically, new bad data discovers a small
  functionality-bug in the main iterative cycle. 
