Two-Node Camera

Imagine you had a real camera in your hand and you want to film a person standing in front of you.  

A single node camera works just like this.  You can move around in the world to change the position of the camera, and you can rotate the camera with your hands to point it in different directions.  You can use the zoom lens of your camera to zoom into the subject.  

Now, imagine you have a tight string tied to your camera lens, and it extends out to a point in space.  Maybe it's attached to the person you're filming, or maybe its just floating at a point in space you've chosen.  But when the position of the string changes, it drags your camera lens around with it.  The camera lens is always dragged by the string when it moves around. If the string is attached to the person, when you move the camera around the camera lens is dragged by the string so the lens is always looking at the person.

That's how a two node camera works.  It has all the capabilities of a one node camera, but it also has a Point of Interest (the other end of the string).  When the POI changes, the camera follows it, because the camera always wants to point toward the POI.  You can use the rotation of the camera to offset this, but the camera will still follow the POI's position, even if the camera is rotated so you can't see the POI.

You may want to select the camera in your timeline and then go to layer/transform/auto orient and see the way the camera behaves if you change from point of interest to path or none. These settings may be helpful for what you're asking.