On combining points and lines in an image sequence to recover 3D structure and motion

Abstract

The authors consider the problem of combining the use of points and lines in a sequence of images to recover the 3D structure and motion. They determine the configurations that yield a resolvable system of equations, treating the cases of two and three views in more detail. For the case of two views, they show that lines are not needed, as they do not contribute to the obtaining of a resolvable system of equations. For the case of three views, the authors list the few interesting resolvable configurations. They also indicate how to obtain rigidity-based equations using structural invariance. Finally, they show why combining points and lines can be particularly useful when occultation occurs.<>