FLEXIBLE 3D MODELS FROM UNCALIBRATED CAMERAS

We describe how to build statistically-based flexible models of the 3D structure of variable objects, given a training set of uncalibrated i mages. We assume that for each example object there are two labelled i mages taken from different viewpoints. From each image pair a 3D struc ture can be reconstructed, up to either an affine or projective transf ormation, depending on which camera model is used. The reconstructions are aligned by choosing the transformations which minimise the distan ces between matched points across the training set. A statistical anal ysis results in an estimate of the mean structure of the training exam ples and a compact parameterised model of the variability in shape acr oss the training set. Experiments have been performed using pinhole an d affine camera models. Results are presented for both synthetic data and real images.