This paper proposes a generalization of the multi- Bernoulli filter called the labeled multi-Bernoulli filter that outputs target tracks. Moreover, the labeled multi-Bernoulli filter does not exhibit a cardinality bias due to a more accurate update approximation compared to the multi-Bernoulli filter by exploiting the conjugate prior form for labeled Random Finite Sets. The proposed filter can be interpreted as an efficient approximation of the $\delta$ -Generalized Labeled Multi-Bernoulli filter. It inherits the advantages of the multi-Bernoulli filter in regards to particle implementation and state estimation. It also inherits advantages of the $\delta$ -Generalized Labeled Multi-Bernoulli filter in that it outputs (labeled) target tracks and achieves better performance.

The Labeled Multi-Bernoulli Filter

In Bayesian multi-target filtering knowledge of parameters such as clutter intensity and detection probability profile are of critical importance. Significant mismatches in clutter and detection model parameters results in biased estimates. In this paper we propose a multi-target filtering solution that can accommodate non-linear target models and an unknown non-homogeneous clutter and detection profile. Our solution is based on the multi-target multi-Bernoulli filter that adaptively learns non-homogeneous clutter intensity and detection probability while filtering.

http://reza.hoseinnezhad.com/papers/JSTSP_2013.pdf

Robust Multi-Bernoulli Filtering

Algorithm design for parallel implementation of the SMC-PHD filter

This paper addresses the problem of distributed multitarget detection and tracking based on the linear arithmetic average (AA) fusion. We first analyze the conservativeness and Frechet mean properties of the AA fusion, presenting new analyses based on a literature review. Second, we propose a target-wise fusion rule for tailoring the AA fusion to accommodate the multi-Bernoulli (MB) process, in which only significant Bernoulli components, each represented by an individual Gaussian mixture, are disseminated and fused in a Bernoulli-to-Bernoulli (B2B) manner. For internode communication, both the consensus and flooding schemes are investigated, respectively. At the core of the proposed MB fusion algorithms, Bernoulli components obtained at different sensors are associated via either clustering or pairwise assignment so that the MB fusion problem is decomposed to parallel B2B fusion subproblems, each resolved via exact Bernoulli-AA fusion. Third, two communicatively and computationally efficient cardinality consensus approaches are presented which merely disseminate and fuse target existence probabilities among local MB filters. Finally, the performance of these four approaches in terms of accuracy, computing efficiency, and communication cost is tested in two simulation scenarios.

On Arithmetic Average Fusion and Its Application for Distributed Multi-Bernoulli Multitarget Tracking

This paper proposes a novel joint probabilistic data association (JPDA) filter for joint target tracking and track maintenance under unknown detection probability and clutter rate. The proposed algorithm consists of two main parts: (1) the standard JPDA filter with a Poisson point process birth model for multi-object state estimation; and (2) a multi-Bernoulli filter for detection probability and clutter rate estimation. The performance of the proposed JPDA filter is evaluated through empirical tests. The results of the empirical tests show that the proposed JPDA filter has comparable performance with ideal JPDA that is assumed to have perfect knowledge of detection probability and clutter rate. Therefore, the algorithm developed is practical and could be implemented in a wide range of applications.

Joint Probabilistic Data Association Filter with Unknown Detection Probability and Clutter Rate.

Extensions of the SMC-PHD filters for jump Markov systems

The cardinality-balanced multi-target multi-Bernoulli (CBMeMBer) filter is a promising algorithm for multi-target tracking. However, there exists a problem that when the legacy tracks cardinality is big enough, the effect of measurement innovation will be negligible, even if the measurement is close to the target state prediction. Such a problem is shown analytically in this study, and then an improved MeMBer filter has been proposed, which balances the posterior cardinality by modifying the legacy rather than the measurement-updated tracks parameters. The sequential Monte Carlo (SMC) implementation of the proposed algorithm is developed and its performance has been verified by simulation experiments.

Improved multi-target multi-Bernoulli filter

The Gaussian mixture cardinality probability hypothesis density (GM-CPHD) tracker is a promising algorithm for multitarget tracking. However, there are two major problems with it. First, when missed detections occur, the probability hypothesis density (PHD) weight will be shifted from the undetected part to the detected part, no matter how far apart the parts are. Second, when targets are close to or cross each other, the GM-CPHD tracker may fail to discriminate different tracks because the score of each track hypothesis in the traditional method is updated by simply summing the log likelihood ratios (LLR) between successive scans. To solve these problems an improved GM-CPHD tracker is proposed that minimizes the effect of the weight shifting and subsequent estimation errors by a dynamic reweighting scheme and improves the performance of track continuity by a dynamic track management scheme. Simulation results show that the improved GM-CPHD tracker is superior to the traditional methods in both the aspects of target state estimate and maintenance of track continuity so that this improved GM-CPHD tracker will have good application prospects.

Improved Gaussian Mixture CPHD Tracker for Multitarget Tracking

The Lagrangian relaxation algorithm is widely used for passive sensor data association. However, there are two major problems about it. Firstly, the cost function of the algorithm is computed by using least square estimation of the target position without taking the estimation errors into account. To solve this problem, a modified cost function is derived which can reflect the correlation between measurements more reasonably owing to the integration of estimation errors. Secondly, due to the fact that building the candidate assignment tree would take a lot of CPU time, we propose a statistic test based on indicator function with a great improvement of the computational efficiency. Simulation results show that both the correlation accuracy and the computational efficiency of the improved relaxation algorithm are higher than that of the traditional one.

Improved relaxation algorithm for passive sensor data association

When the number of targets is unknown or varies with time, multitarget state and measurements are represented as random sets and the multitarget tracking problem is addressed by calculating the first moment of the joint distribution, the probability hypothesis density (PHD), recursively. The PHD filter has been generalized to the cardinalized PHD (CPHD) filter, which propagates not only the PHD but also the full probability distribution on target number. In this paper, a new particle CPHD filter in the Gaussian mixture framework is presented. The CPHD filter uses a bank of Gaussian particle filters (GPFs) to approximate each Gaussian component and does not require clustering to determine target states. Moreover, Quasi-Monte Carlo (QMC) integration method is introduced to approximating the prediction and update distributions of target states. A challenging passive bearings-only multitarget tracking scenario is used to evaluate the performance of the proposed CPHD filter.

Cheng Ouyang

Papers

Extensions of the SMC-PHD filters for jump Markov systems

Improved multi-target multi-Bernoulli filter

Improved Gaussian Mixture CPHD Tracker for Multitarget Tracking

Improved relaxation algorithm for passive sensor data association

A new Gaussian mixture particle CPHD filter for multitarget tracking