A Multi-Subobject Approach to Dynamic Formation Target Tracking Using Random Matrices

Abstract

Bird flocks are typical group targets with various linear formations and high dynamics due to swarm intelligence. This leads to several problems in traditional multi-subobject group target tracking such as shape model mismatch and false correlations. This paper proposes a multi-subobject approach to dynamic formation target tracking. The algebraic graph theory is introduced to analyze the structure of formation targets, then measurements are clustered and combined into subobjects. Based on the existing random matrix approach, two additional filtering branches, including collective filtering and internal structure filtering, are introduced to achieve the robust tracking performance of formation targets. The KL divergence between the prediction and updated densities of the collective filtering is used to determine the change of formation shape. The correct association between the measurements and the subobjects is realized by the guidance of the internal structure filter. Finally, the effectiveness of the proposed method is verified by the simulation and experimental results.