A Multi-Scan Labeled Random Finite Set Model for Multi-object State Estimation

TL;DR

This paper introduces a multi-scan GLMB posterior recursion method for multi-object state estimation, extending existing filtering techniques to improve smoothing performance in multi-object systems.

Contribution

It extends the analytic GLMB recursion to propagate the multi-object posterior, enabling better smoothing in multi-object state estimation.

Findings

Developed a multi-scan GLMB posterior recursion method.

Implemented an approach similar to the GLMB filter for smoothing.

Enhanced multi-object state estimation accuracy.

Abstract

State space models in which the system state is a finite set--called the multi-object state--have generated considerable interest in recent years. Smoothing for state space models provides better estimation performance than filtering by using the full posterior rather than the filtering density. In multi-object state estimation, the Bayes multi-object filtering recursion admits an analytic solution known as the Generalized Labeled Multi-Bernoulli (GLMB) filter. In this work, we extend the analytic GLMB recursion to propagate the multi-object posterior. We also propose an implementation of this so-called multi-scan GLMB posterior recursion using a similar approach to the GLMB filter implementation.