Geometrically-informed sequential Monte Carlo method for dynamic swarm tracking.

This paper introduces a novel approach for dynamic tracking of swarms of autonomous moving agents (such as UAVs) by leveraging the inherent geometric properties of swarm formations. The method treats the entire swarm as a single target, streamlining the tracking process. The state of the swarm is characterized by the location and velocity of its center, as well as the orientation and geometric parameters of the swarm, which are estimated using the sequential Monte Carlo method. One significant aspect of this paper is the formulation of the likelihood function, which incorporates the geometric information of the swarm and is tolerant of the high rate of false alarms, significantly enhancing the robustness of the filter against common sensor-based challenges in swarm tracking. Likelihood functions are proposed for different swarm formation patterns. Our proposed method is also capable of addressing scenarios where the swarm undergoes multiple time-varying formation patterns. The performance of the proposed methods has been numerically tested. The results indicate that the proposed methods accurately estimate swarm movement, formation, and shape in situations with time-varying formation patterns, even in the presence of high false alarms and missed detections, better than the conventional particle filter.