ICP Relation between reference and aligned

[meakcey]

Hi
I have noticed a strange behaviour while using Fine ICP, maybe it is correct according to theory but I could not understand.
While trying to register two point clouds, when I swap the reference and the aligned clouds there is a nearly %50 RMS difference result occured.
My desired reference selection giving 4.7, swapped is giving 2.2
What is the relation of them in used method?

[daniel]

I bet the coverage / extents of the cloud are different? Mind that all the points of the aligned cloud are used by default, which can result in a very different alignment and RMS (use the final overlap parameter to reduce this effect).

[meakcey]

Thank you very much Daniel
They are getting fit very well, happy to see this result =)

But I did not understand what does it mean and how does overlap affect result?

And lastly, is there any corresponding method in PCL as you know?
There is built in methods for ICP but I have not seen overlap parameter while working on it.

[daniel]

ICP will try to minimize the (squared) distance of ALL the points of the aligned cloud. Therefore depending on where they are, the result will be different. If you change the coverage of the aligned cloud, then there's good change for the result to be different (even if it's very subtle). And if the aligned cloud has points that have no equivalent in the reference cloud (because of holes, etc.) then these points will 'shift' the registration towards themselves. If the difference of overlap is too big, then it's a good idea to tell CC to ignore the farthest points. This is how the overlap parameter is used (you roughly tell CC how many points of the aligned cloud are likely to have equivalents in the reference cloud, and CC will only keep the closest ones).

And I don't know much about PCL sorry! The overlap parameter is something I added quite recently to our ICP method, therefore I don't know if there's an equivalent parameter in other tools/libraries.