It might be a nice side result to use your tech to identify foreground elements automatically to generate the camera motion, without using the 2.5D layering. Then you’d have a more traditional Ken Burns effect. It might not work at picking a specific face out of a crowd, but it would work for a lot of scenarios.

I find Apple’s automatic Ken Burns effect to be way too noisy and almost useless because it zooms in/out from random places in the pictures. When I manually edit a Ken Burns effect myself, there’s a lot less movement, and the movement is so much more meaningful and relevant to the video.