The GMCPHD tracker applied to the Clutter09 dataset

The contribution of this paper is twofold: first, it exposes the tracking community to a dataset previously used in acoustics studies and second, it explores the use of the features in this real dataset in clutter removal. For the latter, the Minimal Redundancy Maximal Relevance (MRMR) technique was chosen for feature selection due to its flexibility on big data; the top features ranked by MRMR are sent to a C4.5 decision tree for classification. Contacts that were not identified as clutter are given to the Gaussian Mixture Cardinalized Probability Hypothesis Density (GMCPHD) tracker. Several metrics show that a very small number of features can be employed for satisfactory tracking performance.