**Spetsakis and Aloimonos
1989 :** The problem of
using feature correspondences to recover the structure and 3D motion of a
moving object from its successive images is analyzed. They formulate the
problem as a quadratic minimization problem with a nonlinear constraint. Then
they derive the condition for the solution to be optimal under the assumption
of Gaussian noise in the input, in the maximum-likelihood-principle sense. The
authors present two efficient ways to approximate it and discuss some inherent
limitations of the structure-from-motion problem when two frames are used that
should be taken into account in robotics applications that involve dynamic
imagery. Finally, it is shown that some of the difficulties inherent in the
two-frame approach disappear when redundancy in the data is introduced. This is
concluded from experiments using a structure-from-motion algorithm that is
based on multiple frames and uses only the rigidity assumption.