Add: morphological filters operate on the z dimension, progressive morph...

[chambbj]

...ological filter can be used to segment ground returns

[jspricke]

HI chambbj, thanks for your contribution(s) :). Can you clean this one up a little? Like squashing all the fixup commits. Have a look at git rebase -i (and do a git push -f later). This would make it easier to review, bisect wouldn't break in case, and your flaws won't show up in the PCL history ;). Thanks!

[chambbj]

Didn't know your stance on rebasing and forced pushes. Done!

[jspricke]

please use ::Ptr instead of makeShared () (saves us a copy ;) ).

[taketwo]

I think it makes sense to change function(s) signature so that the input is provided by pointer. A pointer is needed internally anyways. Would be stupid if the end-user has a shared point cloud, then he uses *cloud to call the function, and the function turns this into another shared cloud.

[jspricke]

Force pushes are ok, as long as you are doing it in your own private branch ;) (and no one is basing work on it). On the other hand, this includes quite a lot of different stuff, so it would be great if you could organize this into logical independent commits (instead of historically ;) ). Maybe one for the filter, one for the segmentation, one for the octree change and for the different tools. This would make it easier to review and we wouldn't remove unrelated patches in case we have to revert it. Sorry for the trouble and thanks for the work!

[jspricke]

@taketwo huh, isn't it the same as in https://github.com/PointCloudLibrary/pcl/blob/master/test/octree/test_octree.cpp#L729

[taketwo]

@jspricke Let me restate my idea. morphologicalDilate() uses Octree internally, which requires the input to be supplied in the form of shared pointer to a cloud. morphologicalDilate() accepts input by reference to a point cloud, so it needs to turn the input cloud into a shared pointer to a cloud (which involves complete copy) before using Octree. Now imagine a situation when the user of morphologicalDilate() already has a shared pointer to a cloud. He will need to dereference it to pass to the function. And the function will create another shared pointer inside. This creates a needless copy.

[jspricke]

@taketwo right (I stopped reviewing this when I saw the makeShared ;) ). So passing a ::Ptr in morphologicalDilate() would be the way to go.

[taketwo]

::ConstPtr to be precise ;)

[chambbj]

Thanks for the feedback @taketwo and @jspricke! I'll break this up into four separate pull requests with the suggested modifications.