A mixture model for pose clustering

This paper describes a structural method for object alignment by pose clust ering. The idea underlying pose clustering is to decompose the objects unde r consideration into k-tuples of primitive parts. By bringing pairs of k-tu ples into correspondence, sets of alignment parameters are estimated. The g lobal alignment corresponds to the set of parameters with maximum votes. Th e work reported here offers two novel contributions. Firstly, we impose str uctural constraints on the arrangement of the k-tuples of primitives used f or pose clustering This limits problems of combinatorial nature and eases t he search for consistent pose clusters. Secondly, we use the EM algorithm t o estimate maximum likelihood alignment parameters. Here we fit a mixture m odel to the set of transformation parameter votes. We control the order of the underlying mixture model using a minimum description length criterion. The new alignment method is illustrated on the matching of optical and rada r images of aerial scenes. (C) 1999 Published by Elsevier Science B.V. All rights reserved.