[Gtk-sharp-list] How to improve on the result of ransac?

Wed, 14 Apr 2004 21:43:38 +0800


	 Thanks all guys that helps me on automatic panorama generation algorithm, especially to Alexandre Jenny and Sebastian Nowozin. Now my program can do two image's ransac homography calculation, :)

     But I found a strange phenomena, that is, the ransac algorithm is not robust enough. Although the ransac can get the homography between two images, the total number of correct corresponding pairs varies every time. Sometimes I can get almost all the correct corresponding pairs, and sometimes not(although there still many correct corresponding pairs). I guess the ransac algorithm is not robust enough in my current implementation becuase the fundamental matrix is calculated by random selected points. In VC++, I use the system time as the random seed. 

     Or the KD tree affects the result? I just apply ANN to my program and do not use BBF.
     Does anyone can give me suggestions and advices? Thanks in advance.