The YouTube movie below from Google explains the concept:
Taking a closer look the data is provided by Immersive Media Corp, a digital video imaging company founded in 1994 captured in what is basically streaming video linked to a GPS. Using an 11 lens camera system (pictured right, see 'What is Immersive Media' for more info) data is streamed in and merged in realtime to provide panoramic video. See the Immersive Demo page for examples.
We used an early unit five years ago to capture panoramas around London, the system is easy to set up, basically the camera is mounted via a suction cup to the roof of a car and the data saved via a small form PC.
Pictured below is the car used to capture the images for Google:
The YouTube movie below highlights the use Google Street View in San Francisco - note the ability to zoom but also the numerous problems with stitching and image matching:
While the Immersive system is indeed impressive it does suffer from problems due to cameras going out of alignment and changing lighting levels between each of the 11 views. As such although it allows the rapid capture of cities - all you need to do is drive each route - the quality sometimes leaves a lot to be desired when compared with the more traditional panoramic techniques.
The image below highlights these issues:
Perhaps of more interest to the panoramic community as a whole is the ability to link in your own views using the recently released KML 2.2 beta specifications. This includes the ability to tile images, perhaps allowing fully immersive, high resolution panoramas in both Google Earth and Google Maps.
You can view the Street View direct from here.