Fuzzy Segmentation of Multiple Articulated Elliptical Curves from Sparse Contour Data

In this paper a method is presented for estimating the parameters of a composite object consisting of several "primitive" objects, that undergo rigid transformations. The objects are fitted to the features that are extracted from the data images. The physical relations between the sub-objects are exploited as constraints on the solution space of the resulting optimisation problem. Particular attention in this paper is given to the segmentation problem. This partitioning of the feature data is performed by a fuzzy segmentation, which is integrated within the framework of the parameter estimation problem. An important feature of the method is that no attempt is made to establish closed-form relations. Instead, bounds on the parameters and on the constraints allow adapting the optimisation to the uncertainty in the data and to the knowledge available a priori. The power of the method is that it can accomodate different primitive curves and constraints with the same general structure Application of the method is in the field of human motion estimation where each of part of the body is modelled by a separate geometric object. The method is illustrate with results on both artificially generated and real feature data.