Create a dense voxel grid from a point cloud

[fvgt]

I got a point cloud and I want to create a voxel grid from it. The straightforward functions that open3d provides, e.g. create_from_point_cloud() works fine, but it just creates voxels from the surface of the objects ( see http://www.open3d.org/docs/latest/tutorial/Advanced/voxelization.html#Voxel-carving ).

So my idea was to first compute the occupied voxel indices from the point cloud:

occupied_voxels = np.unique(((points - np.min(points, axis=0))).astype(int), axis=0)

and store them in a set, so I can efficiently use them later for checking:

occupied_voxel_idx = set(map(lambda x: tuple(x), occupied_voxels))

Then, I create a dense voxel grid, according to the bounding box of the point cloud (the widht, height and depth are from the bounding box of the point cloud)

dense_voxel_grid = o3d.geometry.VoxelGrid.create_dense(
            width=width,
            height=height,
            depth=depth,
            voxel_size=1,
            origin=[0, 0, 0],
            color=[1.0, 0.7, 0.0])

Lastly, I create an empty voxel grid:

new_voxel_grid = o3d.geometry.VoxelGrid()

and iterate over the dense voxel grid and adjust the color, if one of the voxels is occupied:

  for voxel in dense_voxel_grid.get_voxels():
      voxel_idx = voxel.grid_index
      if tuple(voxel_idx) in occupied_voxel_idx:
          voxel.color = np.array([0.7, 0.3, 0.1], dtype=np.float64) 
      new_voxel_grid.add_voxel(voxel)

Normally, the color would be from a color map according to the label , but I exlcuded that part for simplicty. The issue is, that when I draw this new grid, I just get a white screen:

o3d.visualization.draw_geometries([new_voxel_grid])

So my approach does not seem to work. Any idea how I can achieve my goal or why my solution does not work?

[IDoCodingStuffs]

+1. This could be really useful for applications like 3D reconstruction from medical image slices, where surface only reconstruction leads to artifacts where different slices intersect like in the example