Robust Distributed Fusion With Labeled Random Finite Sets

Abstract

Abstract-This paper considers the problem of the distributed fusion of multi-object posteriors in the labeled random finite set filtering framework, using Generalized Covariance Intersection (GCI) method. Our analysis shows that GCI fusion with labeled multi-object densities strongly relies on label consistencies between local multi-object posteriors at different sensor nodes, and hence suffers from a severe performance degradation when perfect label consistencies are violated. Moreover, we mathematically analyze this phenomenon from the perspective of Principle of Minimum Discrimination Information and the so called yesobject probability. Inspired by the analysis, we propose a novel and general solution for the distributed fusion with labeled multi-object densities that is robust to label inconsistencies between sensors. Specifically, the labeled multi-object posteriors are firstly marginalized to their unlabeled posteriors which are then fused using GCI method. We also introduce a principled method to construct the labeled fused density and produce tracks formally. Based on the developed theoretical framework, we present tractable algorithms for the family of generalized labeled multi-Bernoulli (GLMB) filters including δ-GLMB, marginalized δ-GLMB and labeled multi-Bernoulli filters. The robustness and efficiency of the proposed distributed fusion algorithm are demonstrated in challenging tracking scenarios via numerical experiments.

Partial Consensus and Conservative Fusion of Gaussian Mixtures for Distributed PHD Fusion

“…The proposed union-type, conservative fusion and partialconsensus-based distributed GM-PHD fusion can be extended in terms of both the communication protocol and the local filter. In the former, other consensus protocols other than averaging schemes (e.g., diffusion [8], [9], flooding [56]) can be applied, while in the latter, multi-Bernoulli filters [57]- [59] and even particle filter-based RFS filters can be employed based on novel mixture reduction or particleresampling schemes. Trajectory2: k ∈ [43,57] x coordinate (m) y coordinate (m) Fig.…”

Distributed Joint Sensor Registration and Multitarget Tracking via Sensor Network

“…Hence, Theorem 1 shows that the WKLA actually coincides with the GCI fusion rule, originally proposed by Mahler [3] as a generalization of Covariance Intersection to arbitrary densities. The minimal cost J f in (18), which is always nonnegative and vanishes only when all the densities are coincident, is known in the literature as GCI divergence [13], [14]. As discussed in [13], the GCI divergence J f makes it possible to quantify, in a principled way, the degree of dissimilarity among a set of RFS densities within the context of GCI fusion.…”

“…In [35] it has been proved that the resulting WKLA multiagent fusion, also known as Generalized Covariance Intersection (GCI), turns out to be immune to double counting of information and is, therefore, resilient to the presence of loops in the sensor network. The minimum cost associated to the WKLA-fused density is known in the literature as GCI divergence [13], [14]. The GCI divergence provides a sensible measure of the degree of dissimilarity among the set of local posteriors (see [13]), and can therefore be minimized with respect to the unknown drift and/or orientation parameters for sensor registration purposes.…”

“…The minimum cost associated to the WKLA-fused density is known in the literature as GCI divergence [13], [14]. The GCI divergence provides a sensible measure of the degree of dissimilarity among the set of local posteriors (see [13]), and can therefore be minimized with respect to the unknown drift and/or orientation parameters for sensor registration purposes.…”

Multiobject Tracking for Generic Observation Model Using Labeled Random Finite Sets

“…In order to solve these problems, several tractable approximations of multi-object Bayes filter have been proposed successively, namely the probability hypothesis density (PHD) filter [5], [9], the cardinalized PHD filter [10], [11], and the multi-Bernoulli filter [1], [12], [13]. With the recent development of labeled set filters [14]- [22] and their enhanced performance compared to previous unlabeled versions, the study on the FISST-based multi-object tracking has recently turned its focus on the labeled random set filters. Vo et al [14] proposed a class of generalized labeled multi-Bernoulli (GLMB) densities 1 and the relevant tracking filter, the GLMB filter.…”