BMVC 2004, Kingston, 7th-9th Sept, 2004

Factorization-based Hierarchical Reconstruction for Circular Motion
H. Zhong and Y.S. Hung (University of Hong Kong)

A new practical method is developed for 3D reconstruction from an image
sequence captured by a camera with constant intrinsic parameters undergoing
circular motion. We introduce a method for enforcing the circular constraint
in a factorization-based projective reconstruction. This is called a circular
projective reconstruction. Given a turntable sequence, our method uses a
hierarchical approach to reconstructing the objects and cameras, which first
computes a circular projective reconstruction of a sub-sequence and then
extends the reconstruction to the complete sequence. Camera matrix and the
motion parameters, i.e. the rotation angles, are computed iteratively in a way
that minimizes the 2D reprojection error. Thus, an optimal reconstruction is
obtained upon convergence. The algorithm is evaluated using real image
(pdf article)