RE: Surface Types from POint-clouds

From: Anshuman Razdan (
Date: Wed Nov 14 2001 - 17:22:55 EET

        YOu bring up an excellent question. Segmentation of triangular meshes (what you
get from point cloud data) is a hard problem. There are two parts to the
question. One given a set of triangles how to separate them so that each is a
distinct "feature" in it self. The second problem is that once you have
segmented then how to recognize that the surface is sphere, cylinder etc.
Feature recognition from CAD data has been going on for a long time but it is
some what easier problem. The reason being that a cylinderical face in a CAD
data is still "topologically single face" where as in a mesh you can have
hhundreds of trinagles making up that face.

        Having said that -- we have been working on this problem for some time. We have
a good algorithm to solve the segmentation problem. Go to
and under publications look for Watershed based segmentation. There is another
paper in development and I will put it up on the website in next few weeks as we
give it a final revision.

        Now for the recognition problem. If you have classical geometric forms then you
can use differential geometry concepts to classify the segmented surface. For
example cylinder has zero gaussian curvature (cylinder is a special case of a
cone), a torus has a genus of 1, etc. The problem arises that the point cloud
data is always noisy and curvature estimation (which has to be done on this
discrete surface) is a problem. But it can be done.

        There is always the manual method of fitting analytical or implicit surfaces on
parts of the point cloud data. What I mentioned above is an automated method.


Dr. Anshuman Razdan
Director, PRISM
Phone: (480) 965 0483 (Tina)
Office: GWC 574

>> -----Original Message-----
>> From: []On
>> Behalf Of Ching Lee
>> Sent: Friday, November 09, 2001 9:36 AM
>> To:
>> Subject: Surface Types from POint-clouds
>> Hello:
>> Given a point cloud from a surface, is it possible to
>> guess the type of surface: I mean is it possible to
>> classify the surface as clylindrical, conical,
>> toroidal, spherical, etc?
>> After guessing the surface type, is it possible to fit
>> the analytical surface (rather than a free-form
>> surface)?
>> Could somebody please point me to literature on this
>> topic of using reverse engineering data?
>> Thanks in advane.
>> Have a nice day!
>> C K Lee
>> __________________________________________________
>> Do You Yahoo!?
>> Find a job, post your resume.
>> For more information about the rp-ml, see

For more information about the rp-ml, see

This archive was generated by hypermail 2.1.2 : Fri Jan 04 2002 - 09:58:05 EET