Question about registration with ICP

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Question about registration with ICP

frank08
Hi everyone,

I am working on the registration of two sets of cloud lately and I met a problem that I cannot think through.
Take the classic "bunny" for example, I transformed the original bunny cloud with a certain transformation and got the "transformed bunny". Now I am trying to register them. First I use the SAC-IC for initial alignment then use ICP for refinement. But here comes the problem: if I change the original bunny cloud from "source" to "target", I will get different registration result. From the pics below, the Case #2 gets better result.
I suppose the registration is not related to which one is the the "source" and which one is the "target", so I was expecting the same result here. Somehow I am wrong and I don't know where goes wrong. Could someone please help me out? Thanks in advance.
Case #1:




Case #2:



Reply | Threaded
Open this post in threaded view
|

Re: Question about registration with ICP

VictorLamoine
Administrator
Hello,

In case one you seem to have stopped ICP before it converges to a perfect solution, as you are registering equal point clouds, let it run until all euclidian distances are zero.

Also note that the initial SAC-IA alignement are not the same in the two cases, that explains the difference between the ICP alignments.

In general, registering A to B or B to A does not change much the result, but sometimes it has some importance in the tweaking of the parameters!

Bye