Investigate efficient transformation (repositioning, orientation) of captured pointclouds

[jackjansen]

It would be good if we could efficiently move and orient captured pointclouds.

In VRTogether we sometimes have the coordinate space (in virtual world coordinates) where the point cloud is, and which way it is directed: if we ask the user to stand straight and look forward we can use the HMD position (the Unity camera) to get this information.

The first bit of functionality we need is getting the centroid of a point cloud. This could be a method on cwipc_pointcloud objects or it could be meta-information returned by the capturers. The latter is more efficient (we could obtain it while we are enumerating over the points) but the former is more general. Maybe we want both. Or maybe do it like cellsize: the information is stored in the cwipc_pointcloud by the capturer, and if it isn't set the pointcloud method will re-compute it.

The second bit of functionality is that we should be able to apply a 4x4 transformation matrix to the pointcloud. This could be a function (like cwipc_crop and such) returning a new pointcloud, but again it might be more efficient to allow specifying an additional transformation matrix to the capturer (because it could apply it to each point as it enumerates over the points to apply the cameraconfig matrix).

Maybe we should first experiment with this functionality on pointcloud objects, and keep the capturer ideas for later.

@troeggla assigning to you to see if you're interested. If not: assign back to me.

[jackjansen]

The --filter transform option to the cwipc command line utilities would obviously also hugely benefit from this.