Nope 
It’s a little counter-intuitive. I’m just starting to knock an example together, but will need to get back later this morning…
Imagine a point at the center of the cylinder. Look at rotation trace from that point to the ball. This gives the correct line, now extened out a fixed amount, that’s where the camera goes.
The only slightly tricky part is finding the center of the cylinder opposite the ball. But I think if you have segments of the tower as BPs, this will not be a problem and lead to an interesting camera effect.