Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Kernel-Based Ensemble Gaussian Mixture Probability Hypothesis Density Filter

Dalton Durant, Renato Zanetti

TL;DR

The paper tackles multi-target tracking under nonlinear and non-Gaussian conditions by proposing the EnGM-PHD filter, which fuses GM-PHD and SMC-PHD through Kernel Density Estimation to convert a Dirac particle prior into a Gaussian mixture. This yields a posterior intensity $v(\mathbf{x})$ that benefits from both the Gaussian-mixture efficiency and the particle-based nonlinear handling, with convergence guarantees as the component count grows. A key contribution is the uniform-weight KDE-based prior construction and the demonstration that EnGM-PHD reduces to the Ensemble Gaussian Mixture Filter (EnGMF) in the single-target, idealized case. Numerical experiments show EnGM-PHD outperforms GM-PHD and SMC-PHD in a two-target radar scenario, with favorable OSPA performance and competitive computational cost, albeit with some inefficiency due to the state-extraction step.

Abstract

In this work, a kernel-based Ensemble Gaussian Mixture Probability Hypothesis Density (EnGM-PHD) filter is presented for multi-target filtering applications. The EnGM-PHD filter combines the Gaussian-mixture-based techniques of the Gaussian Mixture Probability Hypothesis Density (GM-PHD) filter with the particle-based techniques of the Sequential Monte Carlo Probability Hypothesis Density (SMC-PHD) filter. It achieves this by obtaining particles from the posterior intensity function, propagating them through the system dynamics, and then using Kernel Density Estimation (KDE) techniques to approximate the Gaussian mixture of the prior intensity function. This approach guarantees convergence to the true intensity function in the limit of the number of components. Moreover, in the special case of a single target with no births, deaths, clutter, and perfect detection probability, the EnGM-PHD filter reduces to the standard Ensemble Gaussian Mixture Filter (EnGMF). In the presented experiment, the results indicate that the EnGM-PHD filter achieves better multi-target filtering performance than both the GM-PHD and SMC-PHD filters while using the same number of components or particles.

Kernel-Based Ensemble Gaussian Mixture Probability Hypothesis Density Filter

TL;DR

The paper tackles multi-target tracking under nonlinear and non-Gaussian conditions by proposing the EnGM-PHD filter, which fuses GM-PHD and SMC-PHD through Kernel Density Estimation to convert a Dirac particle prior into a Gaussian mixture. This yields a posterior intensity that benefits from both the Gaussian-mixture efficiency and the particle-based nonlinear handling, with convergence guarantees as the component count grows. A key contribution is the uniform-weight KDE-based prior construction and the demonstration that EnGM-PHD reduces to the Ensemble Gaussian Mixture Filter (EnGMF) in the single-target, idealized case. Numerical experiments show EnGM-PHD outperforms GM-PHD and SMC-PHD in a two-target radar scenario, with favorable OSPA performance and competitive computational cost, albeit with some inefficiency due to the state-extraction step.

Abstract

In this work, a kernel-based Ensemble Gaussian Mixture Probability Hypothesis Density (EnGM-PHD) filter is presented for multi-target filtering applications. The EnGM-PHD filter combines the Gaussian-mixture-based techniques of the Gaussian Mixture Probability Hypothesis Density (GM-PHD) filter with the particle-based techniques of the Sequential Monte Carlo Probability Hypothesis Density (SMC-PHD) filter. It achieves this by obtaining particles from the posterior intensity function, propagating them through the system dynamics, and then using Kernel Density Estimation (KDE) techniques to approximate the Gaussian mixture of the prior intensity function. This approach guarantees convergence to the true intensity function in the limit of the number of components. Moreover, in the special case of a single target with no births, deaths, clutter, and perfect detection probability, the EnGM-PHD filter reduces to the standard Ensemble Gaussian Mixture Filter (EnGMF). In the presented experiment, the results indicate that the EnGM-PHD filter achieves better multi-target filtering performance than both the GM-PHD and SMC-PHD filters while using the same number of components or particles.
Paper Structure (14 sections, 1 theorem, 25 equations, 4 figures, 2 tables)

This paper contains 14 sections, 1 theorem, 25 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background and Motivation
  3. Probability Hypothesis Density Filter
  4. Gaussian Mixture Implementation
  5. Particle Implementation
  6. Kernel-Based Ensemble Gaussian Mixture Implementation
  7. EnGM-PHD Filter Recursion
  8. Ensuring Uniform Weights in the Prior Intensity
  9. Relation to the Ensemble Gaussian Mixture Filter
  10. Numerical Experiment
  11. Model Parameters
  12. Multi-Target Performance Metric
  13. Results
  14. Conclusions

Key Result

Theorem 3.1

(Reduction of the EnGM-PHD filter to the EnGMF) In the special case of a single target with no births, deaths, clutter, and perfect detection probability, the EnGM-PHD filter equations simplify exactly to those of the standard EnGMF from ref:yun2022.

Figures (4)

  • Figure 1: This figure shows the 3-dimensional true trajectories (black) of the two targets and the extracted position state estimates of the compared filters. The clutter is represented by the gray crosses which are Poisson distributed in the red rectangular surveillance region. These are the results of 250 Monte Carlo simulations, overlaid with a single run emphasized to illustrate.
  • Figure 2: This figure shows the true cardinality (black) and the extracted cardinality estimates of the compared filters vs. time. These are the results of 250 Monte Carlo simulations, overlaid with a single run emphasized to illustrate.
  • Figure 3: This figure shows the multi-target accuracy (OSPA) of each filter vs. time. Results are averaged over 250 Monte Carlo simulations.
  • Figure 4: This figure compares the efficiency of the filters and how their number of components relate to their wall-clock times. Results are averaged over 250 Monte Carlo simulations. Using an Intel Core i7 9700K CPU at a base speed of 3.00 GHz and with 16 GB of RAM.

Theorems & Definitions (2)

  • Theorem 3.1
  • proof