Reconstruction of Measurement Matrices for Recovering Shape and Motion from Long Image Sequences.
Ying-jieh Huang, Hirobumi Nishida · 1998
The factorization method for recovering 3D information from image sequences assumes that all feature points selected on the first £rame can be tracked throughout. This assumption is violated, however, if some features disappear or new ones are introduced later. In this paper, based on the factorization method with the paraperspective projection model, we present a method for estimating image coordinates of occluded feature points in order for the estimated locations.to approximate closely their perspective projections. The estimation accuracy is evaluated with synthetic data and some results are presented for 3D information extraction fiom real image sequences. 1