[Insight-users] Estimation of rotation scales

[Luis Ibanez]

Hi Sam,

Here is an heuristic rule for estimating good
values for the rotation/translation scales.

1) Estimate the bounding box of your points
    (in physical units).

2) Take the 3D Diagonal of that bounding box

3) Multiply that by 10.0.

4) use 1.0 /[ value from (3) ] as the translation
    scaling value.

5) use 1.0 as the rotation scaling value.

With this operation you bring the translation units
to the range of rotations (e.g. around -1 to 1).
After that, all your registration parameters are
in the relaxed range of -1:1. At that point you
can start setting your optimizer with step lengths
in the ranges of 0.001 if you are conservative, or
in the range of 0.1 if you want to live dangerously.
(0.1 radians is about 5.7 degrees).


This heuristic rule is based on the naive assumption
that your registration may require translations as
large as 1/10 of the diagonal of the bounding box.


   Regards,


      Luis


------------------
Sam Theisens wrote:

> Hi,
> 
> does anybody know of a way to estimate good rotation scales from the data?
> Currently I am trying to register point clouds using the standard ICP
> algorithm. I estimate the rotation scales from the maximal displacement that
> a rotation around each of the axes can account for. This gives me upper
> bounds to the translation to rotation ratio, sort of. 
> Is this a good way to do this, or are there better ways?
> Thanks in advance for any ideas.
> 
> Sam Theisens
> 
> 
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users
> 
>