Behavior-aware group target tracking for AAV swarms with dynamic structural evolution

Most traditional approaches to multiple unresolvable group target tracking simplify the modeling of group evolution, limiting their effectiveness in scenarios with complex group dynamics. In practical applications, dense group targets, such as autonomous aerial vehicle (AAV) swarms, often exhibit dynamic structural changes including merging and splitting, which pose significant challenges for reliable tracking and behavior interpretation. To address these challenges, this paper proposes a modular, closed-loop framework for group target tracking and behavior recognition (GTTBR). The proposed framework integrates measurement clustering, cluster-to-group association, behavior recognition, state estimation, and track management within a unified recursive process. Variational Bayesian (VB) inference is employed in key stages, including measurement clustering and group state estimation. Within this framework, a behavior-aware extent prediction mechanism is developed, where group behaviors inferred from association patterns are incorporated into the prediction of spatial extents, enabling adaptive modeling of dynamically evolving group formations. In addition, a robust track management mechanism with track-backward correction is introduced to mitigate clustering errors and improve temporal consistency. Experiments on synthetic simulations and real-data measurements after clutter removal demonstrate that the proposed method effectively tracks evolving group structures and provides interpretable descriptions of group behavior evolution.