bundle adj and quality

13 views
Skip to first unread message

Mauro S

unread,
Aug 24, 2022, 9:05:10 AM8/24/22
to Ames Stereo Pipeline Support
Hello all,

Regarding the Bundle adjustment  ip methods ( -ip-detect-method  0=OBAloG, 1=SIFT, 2=ORB.) Is there a general rule which would be better of depends on the images?

And does a larger ammount of ip_matches (in general) guarranties a better dem generation later?

Oleg Alexandrov

unread,
Aug 24, 2022, 11:46:52 AM8/24/22
to Mauro S, Ames Stereo Pipeline Support
These interest point matching algorithms should all work reasonably well, and the default, 0, is good enough. Method 2, ORB, is meant to be much faster and not as precise. These methods are overall not that different, they do some local blur and image gradient and find features based on how big the local change in the image is.

> And does a larger ammount of ip_matches (in general) guarranties a better dem generation later?

More interest point matches make the cameras more self-consistent and the alignment better. Given that the alignment transform and camera positions/orientations have around 6-7 parameters each, more than 100 interest points will likely not add much. It is like fitting a line with N points. When N is big the line does not get so much more accurate after a while. More interest points and better-distributed ones may be useful if you optimize the camera distortion, but that is an advanced topic and not part of regular bundle adjustment. 

--
You received this message because you are subscribed to the Google Groups "Ames Stereo Pipeline Support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ames-stereo-pipeline...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/892ed0c1-2cfd-40e6-817a-6e704deaa9f2n%40googlegroups.com.
Reply all
Reply to author
Forward
0 new messages