Page 2 of 2

Re: ICP best registration with 4 points clouds.

Posted: Mon Sep 08, 2014 3:38 pm
by daniel
Well, in the ICP registration process, the DATA cloud moves relatively to the MODEL cloud. And there's no created cloud, the DATA cloud is simply transformed in its new position (hopefully closer to the MODEL cloud) and renamed with '.registered' as suffix. The MODEL cloud is never modified. You don't get the sum of both (you can still merge them afterwards if you really want this).

Re: ICP best registration with 4 points clouds.

Posted: Mon Sep 08, 2014 3:46 pm
by Luca1988SCENE
sorry but i have to contraddict you. When the process ends on the DB tree appears a new cloud, named xxx.registered with the same numbers of point of the DATA cloud...is there something wrong on the screen ?i don't understand this way.Whaen i do the ICP regstration i don't have to lose points as in CC happened.

Re: ICP best registration with 4 points clouds.

Posted: Mon Sep 08, 2014 4:01 pm
by Luca1988SCENE
i understand that the .registered one is not a new cloud but is the DATA one moved on the MODEL one. but i don't undersand the utility of this process ! Finaly the function merge is a sort of ICP Its self?

Re: ICP best registration with 4 points clouds.

Posted: Mon Sep 08, 2014 4:05 pm
by daniel
But this cloud is the DATA cloud (this is why it has the same number of points ;). It has just been renamed (and in effect it is removed then added back to the DB tree for consistency reason, this is why it appears at the bottom generally).

If you rename your clouds 'DATA' and 'MODEL' before ICP, then after ICP you'll get 'DATA.registered' and 'MODEL'.

Re: ICP best registration with 4 points clouds.

Posted: Mon Sep 08, 2014 4:10 pm
by daniel
When we say 'move', we mean that the cloud actually 'moves' in the 3D space (not that its points are transferred in memory).

See http://en.wikipedia.org/wiki/Point_set_registration.

This is (very) useful when two clouds are not expressed in the same coordinate system for instance, and you don't know the transformation between both systems. This is typically the case when you move the scanner between two positions. You'll get two scans expressed in the scanner referential, but you rarely know precisely how the scanner as been displaced.

It can also be used to make slightly different clouds match as best as possible.

What is funny here is that the registration functions are some of the main reasons why users come to CC ;)

P.S. : the merge method simply copies all the points of the second cloud in the first one