CloudCompare forum

Hi!

I have a question about the automatic registration that can be performed in Cloud Compare.

I have two point clouds (same plot in different months) adquired with a mobile laser scanner and I want to compare them. I would like to align them, so I follow the recommended steps that can be found in the manual (1. register with "Rotate/Translate". 2. Creating a subset of clearly overlapping points with the "Interactive Segmentation tool". 3. Apply the Fine registration (ICP) algorith to the data subset and de "model" point cloud and keep its transformation matrix. 4. Aplly the rigid transformation to the original point cloud with that matrix with "Apply transformation").

Apparently, I obtain a better registered point cloud but when I try to save this point cloud as a new point cloud and use it in R or in other softwares the process them, the registration is not as good as it seemed, and both point clouds do not overlap very well. It seems like when I save the registered point cloud ("File > Save"), the coordinates are not correctly saved.

Also mention that I don't know if I using the best method to register these point clouds. I am scanning a forest plot, so in different months they are quite different and it is not easy to pick equivalent points in both point clouds. I don't know if there is another method that could work better in this kind of environments.

Hope the information is clear!
Thanks a lot.

Normally CloudCompare updates the coordinates directly in memory when you apply a transformation. So at saving time you should have the right values (if the file was properly created).

The only exception is with large coordinates (and Global Shift). I which case you have to make sure the clouds all have the same Global Shift (as you are aligning them in the local coordinate system, and not in the global one). Only the 'Align' tool is able to properly manage the global coordinate systems.

And you are right, you would need to find 'anchor' areas that didn't change between the 2 scans. The ideal case would be once again to use the 'Align' tool with clearly identifiable and fixed points... But I guess that's a challenge in a forest!

First of all, thanks for your answer!

Yes, that may be the problem, considering that when I open the LAS file in CloudCompare it appears a window saying that the coordinates are going to be changed from the original coordinate system to a local coordinate system. I thought that the information needed was stored and used to restore the original coordinates at saving time automatically... Do I have to do something in order to restore the Global Shift?

Nevertheless, I am not sure that this is the problem, as when I process the aligned point cloud in R and load a raster product in QGIS I find that they are, more or less, in the same place (if I see the ortophoto it is the correct plot), but moved from the initial position, not as well aligned as it seemed.

Well, it's just that you should take into account the Global Shift before applying your transformation. It means, you'll have to assume that all points have already been translated of a vector GS, so when you apply a transformation R.P+T, you actually apply R(P-GS)+T = R.P + (T - R.GS) (if that makes sense)