Geometrically-Informed Particle Filters for UAV Swarm Tracking

This paper focuses on challenging swarm tracking applications where a relatively large number of small UAVs need to be detected and tracked as a whole swarm. In practice, due to the small size of the UAVs, they are hard to detect and when using radars, the threshold should be set low to ensure an acceptable probability of detection. However, this results in a large number of false alarms. The common multi-target tracking methods are likely to fail in these scenarios. We present a solution to the problem that exploits the information available for the geometry of the swarm. Our solution is formulated as a low-cost particle filter that treats the entire swarm as a singular entity with a composite state vector comprised of motion variables (such as location, speed, orientation) and geometric formation variables (such as shape and size). The solution is generic but detailed formulation is presented for two types of formations. Numerical results demonstrate that our solution effectively mitigates the effect of numerous false alarms, accurately tracking swarm movements and formations.