Maneuvering, Multi-Target Tracking using Particle Filters

TL;DR

This paper develops advanced particle filtering techniques for underwater multi-target tracking, addressing challenges like maneuverability, data association, and computational efficiency, and compares their performance with traditional methods through simulations.

Contribution

It introduces efficient particle filter-based algorithms for multi-target and highly maneuvering target tracking, including IPPF and MMPF, with improved data association via MC-MMJPDAF.

Findings

Particle filters outperform Kalman filters in non-linear, non-Gaussian scenarios.

Proposed methods reduce computational load with fewer particles.

Simulation results demonstrate superior tracking accuracy of the new algorithms.

Abstract

In this work, we develop tracking and estimation techniques relevant to underwater targets. Particularly, we explore particle filtering techniques for target tracking. It is a numerical approximation method for implementing a recursive Bayesian estimation procedure. It does not require the assumptions of linearity and Guassianity like the traditional Kalman filter based techniques and is capable of handling non-Gaussian noise distributions and non-linearities in the measurements as well as target dynamics. The performance of particle filters is verified using simulations and compared with Extended Kalman Filter. Particle filters can track multi-targets and highly maneuvering targets. However, it has higher computational load. The efficient use of particle filters for multi-target tracking using Independent Partition Particle Filter (IPPF) and tracking highly maneuvering targets using Multiple Model Particle Filter(MMPF) are also explored in this work. These techniques require only smaller number of particles and help in reducing the computational cost. Data association problem exists in multi-target tracking due to lack of information at the observer about the proper association between the targets and the received measurements. The problem becomes more involved when the targets move much closer and there are clutter and missed target detections at the observer. Monte Carlo Joint Probabilistic Data Association Filter (MCJPDAF) efficiently solves data association for the mentioned situation. Due to the inability of the standard MCJPDAF to track highly maneuvering targets, Monte Carlo Multiple Model Joint Probabilistic Data Association Filter (MC-MMJPDAF) which combines the technique of Multiple Model Particle Filter(MMPF) in the framework of MC-JPDAF has been proposed. The simulation results shows the superiority of the proposed method.