A human has a really good vision system. We can easily differentiate the window (for example) from the rest of the environment without problem. Also, we have a pretty good 3D system, and with practice we can process the relative position of objects and use this processed information to control the helicopter/quadrotor.

In this project they are using Vicon to determine the position of the quadrotor and the window. Once that information is processed, the quadrotor is properly controlled and has good enough "feedback", the human control can no longer be a match for the precision of a computer.