This is by no means a revolutionary technique, but more of a trick solution i used to extract a reflection and lighting map.
I hope it may be of use to you.
In this particular case, the studio was given a background plate where a torrent of water/flood was to be added crashing down a tunnel.
No visual fx supervisor had attended the shoot so absolutely no information about the set was available.
As these were shots contained water (a highly reflective/refractive material), it was important to get a good, reasonably accurate image of the surrounding environment for it to reflect - thus sitting the effect into the scene more realistically.
The most common and popular way to get this is of course to either photograph the reflection of the environment/set in a chrome sphere or to use wideangle/fisheye/sphereron camera. Perhaps mutiple exposures of this also to later combine to create an HDRI (high dynamic range image) or 'light probe'.

Since we had nothing except the raw shots/plates, and a tight deadline - this is the method i used to extract quick and dirty information from the background plates.  I should mention that this method isnt applicable to every such situation, but works well for interiors and contained sets.


After the plates had been tracked/matchmoved and matching scene geometry created, i grabbed a frame from the sequence (above) to be used as a camera projection map.
Camera projection is a method commonly used for matte paintings, where a still image or matte painting is projected from a virtual cameras perspective onto corresponding geometry. This provides opportunity to create 3d movement and depth to an otherwise 2D image.

By projecting the frame onto the modelled tunnel geometry from the matched camera, we now have a rough recreation of the set environment in which additional virtual cameras can pan,zoom and rotate. Of course there is some major texture distortion in some areas, particulary towards the end of the tunnel where the frame/texture becomes smaller and so stretches more to fit the geometry.  Photogrammetry techniques could also be used here for more accurate object recreation, though this usually needs more than one image to work correctly.

Below is a screengrab of what this simple scene now looks like.  A second camera was created to move independently and render from - the first camera must remain stationary as this is projecting the image onto the scene. It may be a good idea to freeze/lock this camera to avoid accidently moving it and messing up the projection.

enlarge

Because the tunnel in this scene was similar all the way along - i could also mirror the scene so that the tunnel and projection would extend back past the original camera, creating some set recreation behind the original camera POV.

To recover the reflections, i placed a sphere in the middle of the tunnel and gave it a brazil chrome material. The same reflective shader can be easily created using a raytrace material, but seeing as brazil has one all set up and ready - i just stuck that on.
enlarge

The chrome material on the sphere reflects the environment of course, just as if we had photographed a reference sphere on location.  Now we can place the moving camera close to the sphere and hit render.  
The moving camera can be placed anywhere around the sphere to obtain the desired angles we need.

The movie clip below shows a rendered camera move around the sphere.

play

After grabbing the frame/render i needed, i took it into HDR shop to do a 'panoramic transformation'.
HDR shop is a handy little program used generally for...guess what...making HDR images.  I cropped the image to the edges of the sphere and then by going to - image, panorama, panoramic transformations - a menu pops up displaying settings for the source image and a destination image.  The source image should be set to 'mirrorball' and the destination to 'latitude/longitude'.  Set the image resolution/size and click ok.
The chrome/mirror ball render has now been pulled out into an image that can be used for a spherical environment reflection map.  The image below shows the resulting image/map.

ARRRGGGHHHH