1) Make sure you can build a quadcopter that can hover and carry a payload. This, in itself, is hard.
2) Add a webcam and a single-board computer. This may be a Raspberry Pi, or a BeagleBone Black, or something small and powerful like that.
3) Add an object with a known shape/color to the top of your RC.
4) Write code to recognize the shape in the input stream from the webcam.
5) Depending on where in the picture the shape is, make the copter move appropriately.
6) Realize that this will fail as soon as your RC drives under a table/bush/culvert/other cover.