Estimation of Heavy-Tailed Clutter Density using Adaptive RBF Network

In this paper, a method for estimating clutter density using radial basis function (RBF) network is described. Clutter density is important parameter for data association techniques in single and multitarget scenarios. K-distribution is widely accepted model of heavy-tailed sea clutter, however, estimating its parameters using traditional method of moments (MM) or maximum likelihood (ML) approach require computationally intense task. Instead of this, a non-parametric approach is used (density is directly estimated, based on samples in validation volume of tracked target). During tracking process, returns from target and clutter are clustered using Linde, Buzo and Gray (LBG) algorithm, with fixed number of clusters and minimum distance criterion. Based on representative kernel of each cluster, density is constructed and integrated in Viterbi data association filter that also provides a track quality output. Since densities based under target-present and clutter-present hypothesis are available, corresponding likelihood ratios can be used to further discriminate target from clutter and thus enhance tracking process. Although the method for estimating clutter density is described using single target scenario, it is applicable to multitarget case e.g. using multihypothesis Viterbi filter.