I am comparing photogrammetry generated point clouds with a TLS point cloud.
1. With the point pair picking, is the allignment constrained to just these points? Does the fine (ICP) allignment then look at all the other points? I need to see the difference (if any) of point clouds, compared to the TLS, with different lighting arrays so I believe the pair picking would be most suitable.
2. I have dense point clouds (about 3mm). Should I generate a mesh for my reference cloud (TLS)?
Thanks for providing a great piece of software and relatively easy to use.
Marcus
alignment by picking point pairs
Re: alignment by picking point pairs
1. Indeed, only the picked points/pairs will be used for registration.
And by default ICP only uses a subsampled version of the cloud at each iteration (see the 'Random sampling limit' parameter) in order to speed-up the process. But we could say that almost all points will be used after a certain number of iterations. And if you set a partial overlap, then only the closest points will be used at each iteration.
Eventually, for your process it seems indeed a good idea to use the pair picking based registration. But mind the picking process reproducibility.
2. If the clouds is very dense and you don't expect to get more accurate measures than 3 mm, then you can keep on with cloud-to-cloud distances. You can also use a 'local modeling' strategy to reduce the effect of the compared cloud density (see http://www.cloudcompare.org/doc/wiki/in ... omputation).
And by default ICP only uses a subsampled version of the cloud at each iteration (see the 'Random sampling limit' parameter) in order to speed-up the process. But we could say that almost all points will be used after a certain number of iterations. And if you set a partial overlap, then only the closest points will be used at each iteration.
Eventually, for your process it seems indeed a good idea to use the pair picking based registration. But mind the picking process reproducibility.
2. If the clouds is very dense and you don't expect to get more accurate measures than 3 mm, then you can keep on with cloud-to-cloud distances. You can also use a 'local modeling' strategy to reduce the effect of the compared cloud density (see http://www.cloudcompare.org/doc/wiki/in ... omputation).
Daniel, CloudCompare admin