I am using the Kinect sensor to generate a point cloud of a person.
I need to generate from the point cloud the corresponding set of collision surfaces in NVidia PhysX.
To achieve it i have to smooth and after triangulate it to concave mesh/set of convex meshes.
I've found only 1 smoothing algorithm:

The Moving Least Squares.

To triangulate i've found few algorithms:

Greedy Projection Triangulation
Constrained Delaunay triangulation algorithm.
Ear clipping algorithm

And ready solution witch needs an oriented points as input data.

I also want to do it in realtime(at initialization create mesh and after move parts of it corresponding to real person moves) so i need fast methods.
Also smoothing algorithm must return < 1000 points to achieve low polly meshes.
Can u advise some useful methods?

sounds like an intresting problem, but i think to discuss about it we need more information on the pointcloud
are the points spread in 3d space, or is it just a heightmap like this http://makerbot-blog.s3.amazonaws.com/wp-content/uploads/2012/01/3D-Printing-with-Kinect-.jpg

maybe it's possible to generate a voxel lattice over the pointcloud ... to close the space between points u could use a morphologic closing function (http://en.wikipedia.org/wiki/Mathematical_morphology#Closing)
then generate an octree from the closed pointcloud to do a fast collision detection

I once implemented this algorithm:

http://www.cs.ucdavi...nta/pubs/sm.pdf

and it worked quite well (although it was far from real time).

I haven't implemented other algorithms, so I can't tell you how it compares to other methods. Just do a google search for surface reconstruction, which should give you loads of papers about it.