Multiple-Model Tracking of Dynamic Formations Using Group Structure Maneuver Models With Subobject Splitting and Merging

Tracking the complex shapes of group targets, which provides essential information for situational awareness, is a critical task in group target observation. Traditional tracking methods represent group shapes using a fixed number of elliptical sub-objects and model the shape evolution with one simple dynamic model. However, in practice, group shapes change rapidly due to swarm intelligence, and the prior evolution model is often unknown. These factors degrade shape estimation accuracy, leading to poor tracking performance. In this paper, we propose a robust multiple-model tracking approach to accurately track group targets with complex and dynamic shapes. This approach treats the dynamic changes in group shapes as special maneuvers in group structures. First, a group structure maneuver (GSM) model, based on a multi-ellipsoidal representation, is introduced to describe the dynamic behavior of group shapes. Then, a prediction method employing multiple GSM models is designed to capture the dynamic evolution of group shapes and enhance the adaptability to rapid shape changes. Furthermore, to achieve hybrid estimation across different GSM models, a hierarchical fusion strategy is developed. Specifically, a group structure combination method is derived and integrated into multiple parallel estimators to jointly update the number and states of the sub-objects. The outputs of these estimators are then fused using the weighted Kullback-Leibler Average (KLA) method to obtain a high-accuracy overall estimation. Finally, the superiorities of the proposed method, such as early maneuver detection and high tracking accuracy, are validated through simulation and experimental results.