In my case, I am trying to use existing spherical panoramas like those seen in Google's Street View to figure out how to augment a photograph (or, eventually, maybe even video) taken with a mobile device. By matching the photo and panorama and finding the geometry between the two, it it easy to take a model from the panorama and project it into the photo. I recently did a talk describing this, so you can have a look at the slides here (much of the information is in the notes rather than the actual slides).
Once you can accurately add arbitrary models to your photographs, you have a whole new world of possibilities. I love this concept of looking at historical buildings, allowing you to see what a place used to look like in the good old days. I think there are some other cool possibilities, including games. A scavenger hunt could allow players to tag certain structures (like buildings, signs, etc) as their own, and players encountering those later on would see the first player's marker. Kind of like this tagging application, except mine would actually work ;).