My initial thought was recording the position in each frame.

Obviously after reading it, it seems obvious. And if I were implementing it, I am pretty sure I would have came to the same solution.

But it was not immediately obvious to me from the title.