The Recursive Spectral Bisection Probability Hypothesis Density Filter
Ding Wang,Xu Tang,Qun Wan
DOI: https://doi.org/10.1007/978-3-030-36402-1_5
2019-01-01
Abstract:Particle filter (PF) is used for multi-target detection and tracking, especially in the context of variable tracking target numbers, high target mobility, and other complex environments, it is difficult to detect, estimate and track targets in these situations. This paper discusses the probability hypothesis density (PHD) filtering which is widely used in the field of multi-target tracking in recent years. The PHD filter algorithm can estimate the number of targets effectively, however, existing algorithms does not make full use of particle information. This paper proposes a target state extraction method based on the recursive spectral bisection (RSB) node clustering algorithm, which focus on eigenvector centrality, algebraic connectivity, and the Fiedler vector from the established field of spectral graph theory (SGT). The method makes full use of the geometric distance relationship and the weight of particles to construct the particle neighborhood graph, then use the algebraic connectivity and Fiedler vector obtained by the eigenvalue decomposition of the Laplace matrix, finally extracts the target state from each class of particle group. Simulation results demonstrate that the new algorithm provides more accurate state estimations for multi-target detection and tracking.