[JewishDan18]

The thing that makes the math more complicated is that with the circles, it became a matter of tracking something that was an ellipse (or something that was extremely close to one, close enough that the resolution of the camera didn't really pick up a change). With the rectangles, the problem becomes tracking an (to a degree) arbitrary quadrilateral. To have accurate measures, you need to know the four corners, which is hard to do using the automated NI vision stuff (or at least I am finding it to be).

The first paper you give is really cool, but unfortunately, is an algorithm for finding rectangles in a cooridor (with clear perspective lines). They state that the general perspective-transformed-rectangle finding problem, which we are trying to do, is very hard.