Forum

Mesh from pointcloud

is it possible to instantly 3dscan an entire person via multiple kinects ?

  • visibility: not sure what is happening on the concave parts of the hands/legs
  • multiple kinects: kinect1 so far, but i remember reading about addressing this in a new pack via network? count is probably same as for projection, so 4-5 units for 360
  • aligned: rulr or calibrateprojector?
  • merged: seen it in pointcloud pack
  • meshing: seen it in gallery at least
  • uvs: how about projected textures from the kinect pov ?
  • texturing: from rgb camera, incl depth-rgb displace

so far there is some progress on that frontier. however, all working algorithms have been working from the vantage point of the cam, and hence have aweful discontinuities in mesh generation.

to do what you ask, I believe it necessary to use a (preliminary) center point of the object, good filtering of the pointcloud, and to finally wrap, distort and tesselate a cylindrical (or better, a spherical) mesh around it.

to answer a few things in detail:

  • this is not as simple as convex/concave. an arm is concave in nature, but it might still be even more complicating, because it can curve in itself and hide stuff from all cameras. the human eye can usually calc around this, but to do so in gpu is entirely different.
  • readbacks and zeromq is sufficient to stream&unify lots of kinects into one pointcloud. intolight in full effect!
  • rulr is great to calibrate odd canvases. if you just want to join multiple physical kinects with particular physical placement into one virtual space, have a look into the girlpower of dx11-pointcloud. this can be done manually or patched to fully tag-based automatik, depending on your needs
  • adding manual uv into the pointcloud is certainly possible. maybe edit the fxh definition to do so. again, the center point might be key to do so more usefully than the pov of kinects ever can.

i think there is chance to do this, if u create an sdf from each kinect, and then use marching cubes to get trinagle mesh, however this means also u need a pixel perfect aliment…