Allign two clouds by picking (at least 4 pts) equivalent pts with unexpected result?

[ray1234]

Hi , I get some raw data, and I do some proccess.
Step1:
I first use "Allign two clouds by picking (at least 4pts) equivalent points" with Rotation "XYZ" turning on "Tx" "Ty" only.
Step2:
I found that x y are still have some difference to what I expect. So I pick to "allign points" again and use the same set of "reference points" as before.
This time, I choose Rotation XYZ turning on Tx Ty with adjust scale.
Step3:
x y seems to be good enough. I want to move z...
Generally, I pick 5 points, the z in the generated and modified cloud is about move up about +46 metres.
Please find the attached photo for my expected calculation of z increment work.


However, I got trouble...
this is the result


I don't know why the calculation will do a increment of 45.581 in z but not at least 45.90414 or about 46.0 ?
Step4:
Shift+click to get 5 points again... those points are really have 0.5m less than actual reference point in z value.

Step5:
I attempt to use those 5 points to just make a increment in z value again by using "Allign two clouds by picking (at least 4pts) equivalent points" again.
I don't know why it will get a matrix with all z = -0.121 but not +0.3~0.5 ...
Please look at the photo attached... it get a negative values make me stick here....

[daniel]

Which version are you using?

And then, these rotation and translation "filters" are a little bit tricky because they are indeed 'filters' of the ideal 3D registration matrix. They may not be super well defined from a mathematical point of view, hence not always giving the real optimum transformation (if I remember well).

Maybe this could be improved. Can you maybe share your data with me? (admin [at] cloudcompare.org).

[ray1234]

I am using cloudcompare 2.9.1