Swarm-enhanced labeled multi-bernoulli filter for dynamic swarm tracking
Tharani Thathsara Rajapaksha, Amirali Khodadadian Gostar, Alireza Bab‐Hadiashar, Nida Ishtiaq, Reza Hoseinnezhad · Aerospace Science and Technology · 2025
• Swarm-enhanced LMB filter for dense UAV swarm tracking in heavy clutter. • Sequential Monte Carlo swarm-state estimator fused into LMB likelihood. • Geometry-aware ellipse and V-shape formation models guide data association. • Handles time-varying swarm formations via Jump–Markov formation switching. • Up to 28 % OSPA(2) and 34 % cardinality error reduction over standard LM. Tracking individual agents within dense and dynamically maneuvering Unmanned Aerial Vehicle (UAV) swarms poses a significant challenge, particularly when agents operate in close proximity and measurements are often missed and prone to high false-alarm rates. Traditional multi-target filters frequently struggle under these conditions due to ambiguous data association leading to reduced tracking accuracy. To address this, we propose a swarm information incorporated tracking framework that jointly leverages swarm level dynamic and formation related information to enhance tracking accuracy of individual agent within the swarm. The key idea is to estimate swarm-level motion and formation parameters using the Sequential Monte Carlo (SMC) swarm tracking approach by modeling the swarm as a single entity and integrate this information into the Labeled Multi-Bernoulli (LMB) filter to improve individual trajectory estimation of UAVs. This joint formulation improves measurement association, reduces identity ambiguity among closely spaced UAVs, and increases robustness to clutter and missed detection. The performance of the proposed method is evaluated on common swarm formations, such as the elliptical shape, V-shape, and swarm with multiple-varying formation patterns under conditions of imperfect detections and high clutter. Tracking performance is quantified using the Optimal Sub-Pattern Assignment distance for a set of tracks (OSPA (2) ) metric, demonstrating up to an 28.1 % reduction in tracking error and an 33.8 % improvement in cardinality estimation compared to the standard LMB filter. These results highlight the effectiveness of integrating the swarm information to strengthen individual UAV tracking within the swarm, providing a robust framework for real-time monitoring.