Collect (unaccepted) user input events and map them to 2D screen.
The architecture has to be extendable to 3d.
Later it has to be possible to map the events also to 3D objects.

This probably has to be configurable, since users might also want to record all events in some cases.