Robust estimation of motion, structure and focal length from two views of a translating scene

We present a simple algorithm to recover focal-length, motion parameters and three-dimensional structure of an object translating along an unknown direction. The algorithm can be useful for inspection tasks for conveyor-belt systems. In order to obtain a fully metric reconstruction our algorithm uses two non-linear constraints easily available in a real setup. The algorithm is made robust performing a preliminary outlier detection. © 1999 Elsevier Science B.V.