Fixed overlay (depth test disabled); removed meaningless mean error visualization; changed some function and variable naming.

Made a parameter for the threshold of correspondence rejection. Added a very basic widget to provide some feedback for point cloud to virtual world registration.

Some minor performance optimization (which were made in our development software for the paper and are now transfered into this UE4 project).

Further changes will follow.

Calculated the grid mid after the grid node (hole center) rejection is done for a more stable grid mid when the grid nodes are very noisy.

Note: The final matrix is still calculated via the TransformationEstrimationSVD function (to get a rigid transformation matrix). There are only some correspondence points filtered out previously.

- Fixed FTransform (GridDetector.cpp: l.1393 - I'm wondering how the first demo try worked so well with this...)
- Initialized FVector with zeros (GridDetector.cpp: l. 1264, in rare cases this seemed to be not 0).
- Fixed Illegal Memory Access in rare cases (GridDector.cpp: l. 963, 1246).

Separated the grid detection and registration in an extra thread so that the kinects can run at 30fps and the time difference between the used point clouds is smaller.

- Using only the centers of the grid (in case it is not ensured that each Kinect sees the grid from the same side).
- With all the centers of the holes of the grid for faster registration, if it is ensured that the grid is always seen from the same side (e.g. Kinects on the ceiling and look down).

Currently, the images in the sample project are not synchronized (two threads, there can be a delay of multiple images between both kinects!), so the grid must move extremely slowly as it is currently implemented. This will be fixed soon.