Regarding the alignment of the images, it should be possible to use the accelerometers to guide the user of where to place the camera while taking the second shot. I can imagine an application that uses arrows instructing the user in what direction to move and rotate the camera and when ready takes the snapshot.

A more interesting 3D application would be to instruct the user to take multiple images of an object and then build a 3D polygon file out of it, e.g. in the PLY file format.