Finding feasible parametr sets for shape from silhouettes with unknown position of the viewpoints

Abstract

Reconstructing 3D shapes from 2D silhouettes is a common technique in computer vision. It requires knowing the position of the viewpoints with respect to the object. But what can we say when this information is not available? This paper provides a first insight into the problem, introducing the problem of understanding 3D shapes from silhouettes when the relative positions of the viewpoints are unknown. In particular, the case of orthographic silhouettes with viewing directions parallel to the same plane is thoroughly discussed. Also we introduce sets of inequalities, which describe all the possible solution sets and a paving technique to calculate the feasible solution space of each set.