Unsupervised track classification based on Hierarchical Dirichlet Processes

An unsupervised track classification approach is applied to sonar multistatic multitarget tracking. Appropriate discriminative and aggregative features are derived from beamformed and normalized matched-filtered data as recorded from a linear array towed behind an AUV. A clustering algorithm based on Hierarchical Dirichlet Processes is proposed for unsupervised classification of tracks. Overall improvement of target tracking is demonstrated via the Optimal Subpattern Assignment metric.