Effectiveness of Bayesian filters: An information fusion perspective
Introduction
Dynamic state estimation has been a long-standing and vibrant area of research concerned with the sequential process of estimating a/multiple state(s) evolving over time based on noisy observations. It is the core of many fundamental problems including positioning, tracking, econometric forecasting, adaptive control, etc.
A “naïve” estimation solution is to infer the state directly from the noisy observations received in discrete time instants, hereafter referred to as the observation-only (O2) inference, which will be addressed in this paper. This is a computationally fast estimation method, providing accuracy that is completely dependent on the observation noise regardless of the state process (for which there is, therefore, no need to model it).
In contrast to the straightforward O2 inference, which provides only the point state-estimate, the prevailing solution that has been most investigated is to model the system as a hidden Markov process and employ a recursive estimator to statistically fuse the observations with models in real time. In this case, a two-step estimation paradigm must be adopted, including model identification based on data and filter design based on the identified model [41]. The optimal recursive state estimator in the Bayesian sense requires the complete posterior density of the state to be determined as a function of time. The posterior probability density function (PDF) can be analytically computed only for linear systems with additive Gaussian noises for which the known Kalman filter [24], [25] gives the optimal estimate (and some other special cases [9]). In the general case of nonlinear system or/and non-Gaussian noises, it is impossible to compute the exact form of the posterior PDF; instead, one has to resort to some form of approximation which can be parametric (e.g. Gaussian filters or Gaussian sum filters), non-parametric (e.g. Monte Carlo methods) or a mixture of both. An astonishing surge of various recursive filters/smoothers has been witnessed since [24], [25].
These recursive estimators, which have the Bayesian paradigm as the theoretically most elaborated base [26], perform well as long as the models used are accurate, having few disturbances, and that the approximation (required in nonlinear systems) is insignificant. Ideally, an optimality (e.g. Cramér–Rao lower bounds, CRLB [14], [51], [57]) can be reached if the physical world and the assumed model coincide perfectly. However, in most practical problems, accurate knowledge of the state process model (and noises) is often missing. The model of a real process may differ from the assumed model or the best available model for that process, leaving a difference we refer to as modeling error.
It has been well acknowledged in literature since at least [13], [21], [22] that modeling errors (and significant disturbances) can easily cause significant performance deteriorations or even failures of filters. Therefore, dealing with modeling errors has been a fundamental problem. This, however, is not a problem for the O2 inference as it is free of state process modeling. For recursive estimation, a large variety of strategies have been proposed to enhance the filtering performance including model assessment [11], adaptive filtering (e.g. [19], [34]), robust filtering (e.g. effective characteristics [7], particularly including detection and treatment of uncertain noise [45], outlier [38], abrupt motion [36], asynchronous observations [47], [40] and colored noises [53]), “direct” filtering [41], variable rate filtering [15] and finite impulse response filtering [29], just to name a few. Similar issues occur in Bayesian smoothers and predictors [1], [6], [18], [48] as well as other recursive estimators e.g. optimization-based estimator [27], [42], [46]. The situation will be much more complicated in the multi-target case of cluttered environments, see e.g. [3], [30], [31], [55]. We do not intend to detail these in this paper. However, we would like to point out that:
- (1)
While considerable efforts have been devoted to developing sophisticated recursive filters, the general effectiveness of these filters has remained elusive. Simply stated, it is rare to be asked whether the use of a filter will pay off when modeling errors (including outlier noise) occur or when too much approximation is triggered. This is primarily because a clear definition of the effectiveness for general filters is still missing. Such a definition would require a clear, efficient and engineer-friendly benchmark that is qualified to assess all filters in a consistent manner. The same holds for the work on smoothers and other recursive estimators.
- (2)
It has been demonstrated that the Bayesian inference can behave very badly if the model under consideration is erroneous e.g. [18]. More specifically simple deterministic methods outperform the Bayesian filter in a type of finite-state estimation [44] even when the model is properly set up. In any case, the quantitative analysis of the failure of filters is missing. This paper will thoroughly demonstrate that the O2 inference can outperform recursive filters in certain situations, thus indicating that filters do not always pay off.
In this paper, two primary contributions have been made with regard to these fundamental issues.
- (1)
The O2 inference is established as a benchmark, a bottom line, to assess the effectiveness of recursive estimators, including the Bayesian filter, where an estimator is defined as effective only when it can at minimum outperform the O2 inference on average in accuracy. For a nonlinear observation function, a bias is noticed in the O2 inference and, consequently, a Monte Carlo sampling-based debiasing approach is proposed.
- (2)
The effectiveness of the Bayesian filter of the prediction–correction format is quantitatively investigated from the information fusion perspective, and examples are evaluated on classic filtering models via simulation. Both theoretical studies and simulation results show that the O2 inference can easily outperform the filters in certain situations, more so than expected. This deserves particular attention for the application of any filter.
The remainder of the paper is organized as follows. The basic idea of the Bayesian filter and the O2 inference is given in Section 2. Section 3 investigates the effectiveness of the recursive filter from the general perspective of information fusion while Section 4 presents simulation results based on three representative problem models to demonstrate the theoretical findings. We conclude in Section 5.
Section snippets
A brief review of Bayesian filters
The dynamic state estimation, also referred to as the filtering problem, is generally modeled in the state space where the system being modeled is assumed to be a Markov process of hidden state. This can be formulated as a state space model (SSM) that is comprised of a state process equation and an observation equation as follows where t indicates the time instant, xt denotes the state vector, yt denotes the observation (also called measurement) vector, and ut and vt
Probability of filter benefit
For simplicity, both the prior xp and the O2 inference xo are assumed to be subject to Gaussian in the 1-dimensional state space, either biased or unbiased with regard to the true state xT i.e., . Here, we omit the reasons that cause the bias (to the prior or to the O2 inference) and are only concerned with how the bias that once occurred will affect the filtering result in different situations. The Bayesian filter fuses p(xo) and p(xp) obtaining the posterior
Simulations
In this section, we will investigate the effectiveness of several known (extensions of) Kalman filters and particle filters based on two popular one-dimensional state space models, one with Gaussian state process noise and the other non-Gaussian, and a representative maneuvering target tracking case.
Conclusions
The observation-only (O2) inference is a straightforward, and probably the simplest, solution for dynamic state estimation. We have elaborated this method systematically and proposed a Monte Carlo sampling solution for unbiased nonlinear implementation. While the posterior CRLB provides a lower bound on the mean-square error of any “unbiased” estimator of the random parameter, the O2 takes a more practical approach by setting a higher bound on the mean error of any “effective” estimator
Acknowledgments
The authors acknowledge the insights of Prof./Dr. Yu-Chi (Larry) Ho, Huimin Chen, Xiao-Rong Li, Miodrag Bolicć, Petar Djuricć, Mahendra Mallick, Quan Pan, etc. on this work and the language checking by Deanna Garcia. This work is partly supported by European Commission: FP7-PEOPLE-2012-IRSES (ref. 318878) and MSCA-RISE-2014 (ref. 641794) and National Natural Science Foundation of China (No. 51475383) and Tiancheng Li's work has been supported by the Excellent Doctorate Foundation of
References (57)
- et al.
Gaussian filtering and variational approximations for Bayesian smoothing in continuous-discrete stochastic dynamic systems
Signal Process.
(2015) Gaussian sum approximations in nonlinear filtering and control
Inform. Sci.
(1974)- et al.
Identification of process and measurement noise covariances for state and parameter estimation using extended Kalman filter
J. Process Control
(2011) Recursive estimation of high-order Markov chains: approximation by finite mixtures
Inform. Sci.
(2016)- et al.
A receding horizon unbiased FIR filter for discrete-time state space models
Automatica
(2002) - et al.
Fight sample degeneracy and impoverishment in particle filters: a review of intelligent approaches
Expert Syst. Appl.
(2014) - et al.
Refined particle swarm intelligence method for abrupt motion tracking
Inform. Sci.
(2014) - et al.
Multiple object tracking with partial occlusion handling using salient feature points
Inform. Sci.
(2014) - et al.
Nonlinear Bayesian state estimation: a review of recent developments
Control Eng. Practice
(2012) - et al.
Robust weighted fusion time-varying Kalman smoothers for multisensor system with uncertain noise variances
Inform. Sci.
(2014)
Improved robust finite-horizon Kalman filtering for uncertain networked time-varying systems
Inform. Sci.
Advanced point-mass method for nonlinear state estimation
Automatica
Tracking and Data Fusion: A Handbook of Algorithms
Decorrelated unbiased converted measurement Kalman filter
IEEE Trans. Aerosp. Electron. Syst.
Diffusion strategies for distributed Kalman filtering and smoothing
IEEE Trans. Autom. Control
Intrinsically optimal Bayesian robust filtering
IEEE Trans. Signal Process.
Indoor tracking: theory, methods, and technologies
IEEE Trans. Veh. Technol.
Exact finite-dimensional nonlinear filters
IEEE Trans. Autom. Control
Exact particle flow for nonlinear filters
Assessment of nonlinear dynamic models by Kolmogorov–Smirnov statistics
IEEE Trans. Signal Process.
Mean and variance of implicitly defined biased estimators (such as penalized maximum likelihood): applications to tomography
IEEE Trans. Image Process.
Treatment of bias in recursive filtering
IEEE Trans. Autom. Control
A fresh look at Bayesian Cramér–Rao bounds for discrete-time nonlinear filtering
Models and algorithms for tracking of maneuvering objects using variable rate particle filters
Proc. IEEE
Novel approach to nonlinear/non-Gaussian Bayesian state estimation
IEE Proc. F Radar Signal Process.
An Introduction to the Kalman Filter
Adaptive Filtering and Change Detection
Cited by (62)
Chernoff fusion using observability Gramian-centric weighting
2024, Information SciencesA multi-source information fusion model for outlier detection
2023, Information FusionAdaptive Bayesian filtering based restoration of MR images
2021, Biomedical Signal Processing and ControlCombining pre- and post-model information in the uncertainty quantification of non-deterministic models using an extended Bayesian melding approach
2019, Information SciencesCitation Excerpt :There are many approaches in information fusion based on different uncertainty theories, such as Dempster-Shafer evidence theory [33], possibility theory [5], fuzzy set [6] and rough set [37]. Among all in-used information fusion methodologies, Bayesian-based approaches have drawn extensive attentions due to the ability to combine both subjective experts judgements and objective test data in a principled way to perform inference [21,39]. To integrate all available information in reliability analysis, Johnson et al. [17] developed a full-Bayesian approach for a multi-level system.