If the FMV's have characters in them then finding the location of things is easier (as they have a sort of walk mesh)
The background images ALL have walk meshes, I suggest working from that perspective. It is entirely feasible to create 3d content that is rendered to the FF7 data format for backgrounds for each scene. It might be slow and complex but you can then make the output content on any scale. I use POV for a lot of stuff like that. I gave up using Open GL because I would get caught up in the OpenGL idiosyncrasis instead of working on creating the 3d data sets to be viewed. The later is what I wanted to do not learn open GL. However OpenGL turned out to be too much fun (go figure). Erstwhile reality is I just need to extract data to mesh data for POV.
Back to the Avalanche project, good fortune.
Redoing the movies is possible, if you have the background scene for the first background in the game done in 3d then you have a good portion of the opening movie done.
Things you would need to make are the Shinra building (needed for a bunch of backgrounds) the reactors (you need those for a few of your backgrounds anyhow). Bascially as you make data sets and styles they can be reused in animated parts just as easily. Hand drawn although potentially easier is not scalable like 3d data.
if you aren't sure what I'm talking about try
povray.org. Lots of good stuff you can do with it.
Cyb