Hi.
I'm trying to register two clouds of points of two objects sampled with an instrument set to a vertical scanning resolution of 450nm and lateral resolution of 7.8 micrometer.
The two clouds are composed of about 2-3 millions of points and they represent two objects similar to two cylinders (they are two drill bits) with diameter of 2.3 mm and length of 4 mm showing very "small" differences.
I'd like to have some generic informations about the selection of parameters and options to register both togheter.
Have you got some suggestions on it?
Thanks
Ale
Information about registration parameters
Re: Information about registration parameters
Well, first if you have not done so already, I would recommend you to read http://www.cloudcompare.org/doc/wiki/in ... gistration (and especially the description of the 'Align' tool - http://www.cloudcompare.org/doc/wiki/in ... itle=Align - and the 'Fine registration' one - http://www.cloudcompare.org/doc/wiki/in ... ?title=ICP).
Then if you use the 'Fine registration' tool, you'd better beware that the ICP algorithm requires some non-flat geometry to lock the two clouds together and converge. For instance it can't work on perfectly flat clouds or even on perfect cylinders (as one rotation would be unconstrained).
Assuming you have some 'features' that will let ICP converge, then you should mainly set the 'final overlap' parameter correctly (if the clouds have the same extent in both case, then it should be 100%). And you can increase the 'random sampling' limit to make sure that the algorithm doesn't miss those 'features' if they are very small compared to the global shape.
Don't hesitate to send me the clouds if you want more directions (cloudcompare [at] danielgm.net).
Then if you use the 'Fine registration' tool, you'd better beware that the ICP algorithm requires some non-flat geometry to lock the two clouds together and converge. For instance it can't work on perfectly flat clouds or even on perfect cylinders (as one rotation would be unconstrained).
Assuming you have some 'features' that will let ICP converge, then you should mainly set the 'final overlap' parameter correctly (if the clouds have the same extent in both case, then it should be 100%). And you can increase the 'random sampling' limit to make sure that the algorithm doesn't miss those 'features' if they are very small compared to the global shape.
Don't hesitate to send me the clouds if you want more directions (cloudcompare [at] danielgm.net).
Daniel, CloudCompare admin