Many years ago GI in rendering wasn’t as fast and easy to use as it is today. It was indeed possible but for animations you had to pre-bake indirect illumination as much as possible. One way of doing this was to place a bunch of cameras at good positions before shooting rays into the scene. Generally this sounds easy but the interesting question was the following: What are good camera position? To save render time you had to use as few cameras as possible but at the same time it was necessary to cover as much as possible of the scene. Basically this is a visibility problem and exactly what the classical art gallery problem is all about. There exist various algorithms which could be used in different ways. Some are listed here:
A fairly simple and efficient algorithm is described in a paper by Michael Batty and Sanjay Rana. It’s closely related to spatial analysis and since I already had some experience in this field it was finally the algorithm I implemented in Houdini.
M. Batty, S. Rana: The automatic definition and generation of axial lines and axial maps