BundleAdjustTwoViews scale ambiguity + BundleAdjustTwoViewsAngular usage

Skip to first unread message

Horst Wildenauer

Sep 15, 2018, 4:41:14 PM9/15/18
to Theia Vision Library
Hello Chris,

I have two questions regarding your bundle_adjust_two_views implementation. Maybe you can help me out here.

.) BundleAdjustTwoViewsAngular implements Oliensis approach, which you seem not to use anywhere in your code. What are your practical experiences - is it performing worse than  BundleAdjustTwoViews in terms of accuracy?
.) Gauge freedom in BundleAdjustTwoViews: I cant find where you fix the scale of the problem. In the code I see that you fix the first camera, but cant see where the scale of the translation of the second camera is fixed. (You do this in
in BundleAdjustTwoViewsAngular)

I hope my questions arent too stupid. I have no working build (forced to use windows) and relied on just reading the associated Code.

Cheers and thanks in advance,


Chris Sweeney

Sep 21, 2018, 3:16:24 PM9/21/18
to Horst Wildenauer, Theia Vision Library
Hi Horst,

I found that that angular approach did not work as well. I never dug into it too much.

The gauge freedom is not fixed in the 2-view bundle adjustment, but the translation is renormalized upon writing back out. This is because BA will take you to a local minima and so the translation will not drift too far away from unit-norm and it keeps the optimization simple. If the translation scale dramatically changes then the 2-view BA is not well constrained anyways and should not be used.

You received this message because you are subscribed to the Google Groups "Theia Vision Library" group.
To unsubscribe from this group and stop receiving emails from it, send an email to theia-vision-lib...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Horst Wildenauer

Sep 21, 2018, 8:55:28 PM9/21/18
to Theia Vision Library
Hi Chris,

thanks for the info. I see your point regarding the local minima. How would you gauge what is too much change in scale?

Ps: In the meantime I coded both versions and my preliminary results confirm your opinion on Oliensis' Approach to be less accurate.

To unsubscribe from this group and stop receiving emails from it, send an email to theia-vision-library+unsub...@googlegroups.com.
Reply all
Reply to author
0 new messages