Taking several pictures of the same location, identify patches of space that represent people and extract them. I should then be able to extract a clean background free of people. I should then be able to apply the different patches of people/objects into the scene either randomly, or to fill the area, taking care to layer the objects in the correct order and also using some parameter to avoid having two objects occupying a virutal place at the same time.
Another thing I can do is to place the people onto another scene, say, from a video game (and video game players into real life). One of the constraints is to try to get the camera angle in the game as close to the one in the scene.