CloudCompare forum

Hi,
I have tested the registration of the software (with register entities) on two ALS (Airborne Laser Scanning) crossing scan lines. It did not work correctly. However I've tried to test it on a part of the point cloud.
For example just a pair of buildings. It works somehow and I could get the transformation and rotation matrices, which are a little bit different for different objects.
As a solution, it could be used to roughly orient the two scan bands, using the obtained parameters. But I don't know if it is possible in this software?
How can I rotate and transform a Point cloud regarding the obtained parameters?

As another question I would be very thankful if you introduce me some documents, which could help me to get involve into the registration techniques of the point clouds.
Best regards
Mossi

Hello,

the ICP algorithm is meant to finely register two roughly registered dataset. So make sure that your dataset are already roughly registered (you can use the graphical transformation tool - the red arrows next to the scissors icon). Also, if the clouds are too flat/smooth, they may "slide". Eventually, if the overlap area is very small relatively to the moving cloud size, you may also get very bad results). If the clouds don't represent exactly the same thing, you may also get strange results (in this last case, the "Enable furthest points removal" can be helpful).

The ICP algorithm is detailed here: http://www.martinkodde.nl/glacier/data/ ... 121791.pdf

In CloudCompare, the output transformation is automatically applied to the "data" cloud and is also displayed in the console. However, there's nothing you can do with it "as is".

There's another tool in CloudCompare ("Align") that is meant to register two clouds starting from any position. But this is still "work in development" and I don't know it very well. The corresponding article is "Four Points Congruent Sets (4PCS) registration algorithm" by Dror Aiger, Niloy J. Mitra and Daniel Cohen-Or.

Hi,

I would like to do something that seems really similar to what Mossi has described.

daniel wrote:In CloudCompare, the output transformation is automatically applied to the "data" cloud and is also displayed in the console.

Does that mean that there is no way to retrieve the full transformation applied to the original point cloud?

I have two similar point clouds of the same object from two different scanners and I want to align one on the other, but one has a much lower point density. I'm able to register them nicely with ICP (Register Entity) after downsampling the biggest cloud and roughly aligning it by hand. I can't directly register the big cloud because it's too noisy. I need to apply the resulting transform to the original cloud separately. The thing is that the transform resulting from the hand-made translations and rotations used to roughly align the clouds is not displayed, so I can't get the full transform. Am I missing something?

Sidenote: It would be useful to be able to select and copy multiple lines in the output :-)

regards,
Vincent

Indeed, the manually applied transformation is not output in the console (but we could easily do that).

Then simplify the way you can get the matrix from the console (you can already copy a single line, can't you?)

And eventually let you rotate any cloud with any matrix (entered in text format just as the console display them... it's shouldn't be a big deal).

I'll try to ship that in the next release!

daniel wrote:Then simplify the way you can get the matrix from the console (you can already copy a single line, can't you?)

Yes, copying a single line works fine, but it becomes an hassle when copying 4 lines matrices repetitively. In fact, you could simply display the matrix in a single console entry with multi-line text (like the output of points selection / distance measure)!

daniel wrote:I'll try to ship that in the next release!

Ok. No pression. Thanks a lot!

Ok, it's available in the last version (02/14/2012).

Works fine for me. Thanks!

And I like the "apply inverse" option!