It's more the latter, it doesn't actually capture the image like a camera would. Instead it captures and records every single light ray and reconstructs the environment based upon that.
That would then allow you to overlay computer generated images onto basically a heightmap that has 100% accuracy to the real world.
Actually, he was right with his first guess. This isn't about capturing film with a heightmap like RGBd cameras do, it's about capturing all light that goes through a volume of space from all angles. If you can fulyl capture all light that enters a volume of space, you can accurately recreate a view from any point in that space. So, in VR I can move my head around, both by turning my head around and moving forward, backward and side to side and accurately see any view.
This does also allow adding CG elements, but it's real strength is being able to capture enough of the light coming through so that you can almost 'teleport' to where the video was recorded.
That would then allow you to overlay computer generated images onto basically a heightmap that has 100% accuracy to the real world.
At least that's what I got from the video.