Kinect pointcloud to mesh


so far I was very pleased with quads 2d and some nice animation features, but now I would like to go into 3D grids and meshs.

Ok, what I want to do, I want transform the kinect point cloud or the depth picture into a low polygon mesh and dye it a little bit.

here are a few examples:

my questions:
is it better to take the point cloud or the depth picture as starting point?

how can I adjust the internal kinect webcam to the depth cam?

maybe someone can show me the right direction.



I played arround with the amazing 3d stuff now, mesh, vertexbuffer… and enabled although my Grundstudium math.

currently I have the problem to do something like 2d spread - Delaunay (2D) - Mesh in 3D. is there a way to do this Delaunay triangles in 3D?

on the other hand I thought I could do the Delaunay (2D) stuff with the x and y coordinates of the pointcloud, it doesn’t look very nice… where is my fault? (129.3 kB)

I sorted the vertices to get a cleaner mesh. perhaps this works to get a cleaner mesh.

one question remains: how can I adjust the internal kinect webcam to the depth data? I saw the thread, unfortunately the openni setup doesn’t work at the moment.

would be nice if someone could give me a few hints on this topic. anyone have experience with this?

or is there a way to calculate the pixel position in the webcam foto from the vertice data?

please, not all at once. no answers?

anyone interested in doing this job for money, feel free to contact me.


I for one am keeping an eye on this thread :) But, because I am sort of where you are, so I can’t be of much help. I mean, using a Pipet to generate some data to apply to a mesh, yeah but that’s not want you’re asking.

there are two problems

_1 to make a cute mesh out of the pointcloud. actually I’am learning a lot in this area, the results getting betting but it’s far away from my expectation.
ok, man wächst mit den aufgaben.

_2 i have done the colorvalue to point transformation in processing with a simple function, works good for the desired distance. now I am trying to do this transformation in v4.

perhaps I give elliots calibration method a try… actually I didn’t get the openni driver to work with v4.

thanks for your answer!


i colored the pointcloud with the pipet. At the moment I am searching a way to get the exact pixel of the corresponding vector. Is there a function to do this transformation?

You can use the Kinect depth data without background as Displace Texture of a Grid.

Check: heightmap-displacement