In the book Multiple View Geometry in Computer Vision Hartley and Zisserman write that given a stereo image pair, the viewed scene can only be reconstructed under projective ambiguity, i.e. is a projective transformation away from the true scene, no matter how many corresponding landmark matches are given.
When using Bundle Adjustment to find extrinsic + intrinsic camera parameters and the 3D coordinates of the scene, how can I be sure that the given values are meaningful? I know Bundle Adjustment is used for e.g. autocalibration, but the computed combination of parameters and coordinates could be wrong and just result in the same image. How is this solved in real applications?