CloudCompare forum

Hi,
this is a general question about understanding how registering of point clouds work - maybe I am missing the obvious.
I understand that in general it should not be possible to distort a point cloud. So there is means in CC to shift/translate, rotate or scale a point cloud. Also, there is several or at least two ways of registration (a) by aligning to point clouds by using atleast 4 points and (b) fine registration. But if there is no distortion of the point cloud, what sense does it make to have different ways of aligning/registration? For example, if I use methode (a) using 4 highly accurately surveyed points, how can this ever been improved by any other methode using more points. Do I not simply get different values for the error of the tie-points I use rather than improve the registration?

your work is much appreciated

cheers

mat

Well, in some cases, you are right, 4 accurate surveying points are better than any other solution. To be more precise, it' particularly true if the scan is very good and made by an undistorted laser scanner for instance. And of course this scan needs also to be dense enough so that you can accurately point your survey points in the scan. In this case, I would say that this is the best method.

But you don't always have the survey points ;). Or if you have some big distortion in your point clouds, then you'd better have much more than 4 survey points and they'd better be well sampled all over the cloud. So if you have a good scan once again, and its is already georeferenced for instance, then it maybe be more efficient/simple to use it to register another cloud (for instance a UAV photogrammetry one).

And of course, if you merging several clouds and you are interested in the global shape of the resulting cloud, then you may prefer to spread the error on the whole cloud and get an 'average' registration that minimize overall gaps (instead of having a very accurately registered scan in some particular points, but with jumps/breaks at the border.

In the end, it all depends on the quality of your data, and what you want to do ;)