Structure and Motion Estimation From Complex Features in Three Views
Tomas Johansson, Magnus Oskarsson, Kalle Åström · 2002
In this paper we introduce the notion of a quiver, which is a feature based on a point and a number of directions. We investigate the constraints different kinds of quivers pose on the camera geometry. In particular we study three types of quivers, having one, two and three directions respectively. For these quivers we investigate structure and motion estimation for two minimal cases. For 1-quivers, three such features seen in three affine views yields a (unique) linear solution. For three 3-quivers in three uncalibrated projective views we get up to twelve solutions, but simulations show that in most cases we get a unique solution. We also study two-quivers seen in three views, where simulations show that such features give more stable estimation of the trifocal tensor, as opposed to only using line or point information.