hello,
I'm trying to overlap 2 meshes to get the RMS, should i make sure that the overlap is 100% to get a correct RMS? when i pick the point of overlap theres a displayed RMS, how much should it be to go on and evaluate the cloud to cloud distance ?
thanks
registration
Re: registration
When performing an alignment by picking several point pairs, the RMS is only computed on the picked pairs (so it's far from being 'exhaustive').
And when performing an automatic alignment with ICP, the 'overlap' parameter is meant to help the ICP algorithm to only take the right points into account (in case the matching surfaces are partial or if the two entities don't actually have the same shape). You need to set the correct overlap percentage to get the right alignment, not to get the right RMS ;)
You should perform the alignment without caring too much for the RMS. Then, you can compute the distances between the two aligned entities and eventually compute the distances distribution parameters (with 'Edit > Scalar Fields > Compute stat. params' - then choose the Gauss distribution). You'll see in the Console the RMS of the whole set of distances.
(you'll need the latest 2.9.beta version)
And when performing an automatic alignment with ICP, the 'overlap' parameter is meant to help the ICP algorithm to only take the right points into account (in case the matching surfaces are partial or if the two entities don't actually have the same shape). You need to set the correct overlap percentage to get the right alignment, not to get the right RMS ;)
You should perform the alignment without caring too much for the RMS. Then, you can compute the distances between the two aligned entities and eventually compute the distances distribution parameters (with 'Edit > Scalar Fields > Compute stat. params' - then choose the Gauss distribution). You'll see in the Console the RMS of the whole set of distances.
(you'll need the latest 2.9.beta version)
Daniel, CloudCompare admin
Re: registration
thank you for the quick reply, so to ensure proper alignment i should keep on selecting points till i approximately think the alignment is proper then compute the point distance?
Re: registration
one last question, when sampling the meshes there are two option to sample in numbers and density, which should be my selection knowing my meshes are scanned objects from white laser scanner. and if number of points how many points should i choose ?
Re: registration
I bet the simplest way is to use the 'number of points' and make several trials until you find a good balance (enough points / density, while not too big ;).
Daniel, CloudCompare admin
Re: registration
when cloud to cloud distance is computed there are 2 values results: mean and standard deviation , which of these values is equal to root mean square error.
Re: registration
None of them :D.
The standard deviation is close to the RMS though (see viewtopic.php?t=2138). You can now also ask CC to compute it afterwards with the 'Edit > Sf > Compute Stat. params' method of the latest 2.9 beta version (the RMS is computed and displayed in the Console when you call this method with the 'distances' scalar field activated).
The standard deviation is close to the RMS though (see viewtopic.php?t=2138). You can now also ask CC to compute it afterwards with the 'Edit > Sf > Compute Stat. params' method of the latest 2.9 beta version (the RMS is computed and displayed in the Console when you call this method with the 'distances' scalar field activated).
Daniel, CloudCompare admin