Swarm-enhanced labeled multi-bernoulli filter for dynamic swarm tracking
Year: 2025
Type: article
Abstract: Tracking individual members of the swarm is highly challenging due to the high density of the swarm, the close proximity of its members, and the dynamic variation in its formation patterns. This paper presents a method for accurately estimating the trajectories of individual swarm members by incorporating swarm-level information into a multi-object tracking filter, specifically the Labelled Multi-Bernoulli (LMB) filter. Swarm information is estimated in parallel with the filtering process using a geometrically informed Sequential Monte Carlo approach that models the swarm as a unified entity, which estimates both the dynamic and formation-related parameters. This integration enhances the robustness against false alarms and missed detections, improving the effectiveness of the filter in real-time swarm tracking. This paper presents the formulation of the proposed framework for two common swarm formations, namely, the elliptical and V-shape, and extends it to handle time-varying formation patterns. The performance of the proposed framework is evaluated through numerical experiments on swarms under false alarms, and the OSPA$^{(2)}$ results indicate a significant improvement in terms of tracking accuracy.
Source: Aerospace Science and Technology
Institution: RMIT University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $4,360
OpenAlex ID: W4417027441