In target tracking and sensor fusion contexts it is not unusual to deal with a large number of Gaussian densities that encode the available information (multiple hypotheses), as in applications where many sensors, affected by clutter or multimodal noise, take measurements on the same scene. In such cases reduction procedures must be implemented, with the purpose of limiting the computational load. In some situations it is required to fuse all available information into a single hypothesis, and this is usually done by computing the barycenter of the set. However, such computation strongly depends on the chosen dissimilarity measure, and most often it must be performed making use of numerical methods, since in very few cases the barycenter can be computed analytically. Some issues, like the constraint on the covariance, that must be symmetric and positive definite, make it hard the numerical computation of the barycenter of a set of Gaussians. In this work, Fixed-Point Iterations (FPI) are presented for the computation of barycenters according to several dissimilarity measures, making up a useful toolbox for fusion/reduction of Gaussian sets in applications where specific dissimilarity measures are required.

In many contexts Gaussian Mixtures (GM) are used to approximate probability distributions, possibly time-varying. In some applications the number of GM components exponentially increases over time, and reduction procedures are required to keep them reasonably limited. The GM reduction (GMR) problem can be formulated by choosing different measures of the dissimilarity of GMs before and after reduction, like the Kullback-Leibler Divergence (KLD) and the Integral Squared Error (ISE). Since in no case the solution is obtained in closed form, many approximate GMR algorithms have been proposed in the past three decades, although none of them provides optimality guarantees. In this work we discuss the importance of the choice of the dissimilarity measure and the issue of consistency of all steps of a reduction algorithm with the chosen measure. Indeed, most of the existing GMR algorithms are composed by several steps which are not consistent with a unique measure, and for this reason may produce reduced GMs far from optimality. In particular, the use of the KLD, of the ISE and normalized ISE is discussed and compared in this perspective.