Tiancheng Li

Review: Fight sample degeneracy and impoverishment in particle filters: A review of intelligent approaches

During the last two decades there has been a growing interest in Particle Filtering (PF). However, PF suffers from two long-standing problems that are referred to as sample degeneracy and impoverishment. We are investigating methods that are particularly efficient at Particle Distribution Optimization (PDO) to fight sample degeneracy and impoverishment, with an emphasis on intelligence choices. These methods benefit from such methods as Markov Chain Monte Carlo methods, Mean-shift algorithms, artificial intelligence algorithms (e.g., Particle Swarm Optimization, Genetic Algorithm and Ant Colony Optimization), machine learning approaches (e.g., clustering, splitting and merging) and their hybrids, forming a coherent standpoint to enhance the particle filter. The working mechanism, interrelationship, pros and cons of these approaches are provided. In addition, approaches that are effective for dealing with high-dimensionality are reviewed. While improving the filter performance in terms of accuracy, robustness and convergence, it is noted that advanced techniques employed in PF often causes additional computational requirement that will in turn sacrifice improvement obtained in real life filtering. This fact, hidden in pure simulations, deserves the attention of the users and designers of new filters.

Algorithm design for parallel implementation of the SMC-PHD filter

The sequential Monte Carlo (SMC) implementation of the probability hypothesis density (PHD) filter suffers from low computational efficiency since a large number of particles are often required, especially when there are a large number of targets and dense clutter. In order to speed up the computation, an algorithmic framework for parallel SMC-PHD filtering based on multiple processors is proposed. The algorithm makes full parallelization of all four steps of the SMC-PHD filter and the computational load is approximately equal among parallel processors, rendering a high parallelization benefit when there are multiple targets and dense clutter. The parallelization is theoretically unbiased as it provides the same result as the serial implementation, without introducing any approximation. Experiments on multi-core computers have demonstrated that our parallel implementation has gained considerable speedup compared to the serial implementation of the same algorithm. A fully and unbiasedly parallel implementation framework of the SMC-PHD filtering is proposed based on the centralized distributed system that consists of one central unit (CU) and several independent processing elements (PEs). Display Omitted An algorithmic framework for parallel SMC-PHD filtering is proposed.All the main calculations of the filter are unbiasedly paralleled.The parallelization obtains theoretically the same result as the serial implementation.Considerable speed-up is gained.

High-speed Sigma-gating SMC-PHD filter

To solve the general multi-target tracking (MTT) problem, an improved Sequential Monte Carlo (SMC) implementation of the probability hypothesis density (PHD) filter called as Sigma-gating SMC-PHD filter, is proposed that updates particles only using the local nearby measurements inside a specified sigma-gate. The sigma-gate is based on the given measurement noise, e.g. 3@s, where @s is the standard deviation of the measurement noise. Correspondingly, a compensation strategy based on the cumulative distribution function of the measurement model is suggested. Eliminating the contribution of measurements lying outside the gate around the particle will highly reduce unnecessary computation and thus improve the overall processing speed. More importantly, this could shield the estimate from interference from the clutter outside the gate giving more robust and accurate estimation. Especially when the clutter density is high, our approach can yield a win-win that is much faster processing efficiency and better estimation accuracy (as compared with the standard PHD filter). This is demonstrated by simulations of the SMC-PHD filters using measurements of range and bearing, respectively.

Monte Carlo localization for mobile robot using adaptive particle merging and splitting technique

Monte Carlo localization (MCL) is a success application of particle filter (PF) to mobile robot localization. In this paper, an adaptive approach of MCL to increase the efficiency of filtering by adapting the sample size during the estimation process is described. The adaptive approach adopts an approximation technique of particle merging and splitting (PM&S) according to the spatial similarity of particles. In which, particles are merged by their weight based on the discrete partition of the running space of mobile robot. Using the PM&S technique, a Merge Monte Carlo localization (Merge-MCL) method is detailed. Simulation results illustrate that the approach is efficient.

A Survey of Recent Advances in Particle Filters and Remaining Challenges for Multitarget Tracking

We review some advances of the particle filtering (PF) algorithm that have been achieved in the last decade in the context of target tracking, with regard to either a single target or multiple targets in the presence of false or missing data. The first part of our review is on remarkable achievements that have been made for the single-target PF from several aspects including importance proposal, computing efficiency, particle degeneracy/impoverishment and constrained/multi-modal systems. The second part of our review is on analyzing the intractable challenges raised within the general multitarget (multi-sensor) tracking due to random target birth and termination, false alarm, misdetection, measurement-to-track (M2T) uncertainty and track uncertainty. The mainstream multitarget PF approaches consist of two main classes, one based on M2T association approaches and the other not such as the finite set statistics-based PF. In either case, significant challenges remain due to unknown tracking scenarios and integrated tracking management.

Approximate Gaussian conjugacy: parametric recursive filtering under nonlinearity, multimodality, uncertainty, and constraint, and beyond

Since the landmark work of R. E. Kalman in the 1960s, considerable efforts have been devoted to time series state space models for a large variety of dynamic estimation problems. In particular, parametric filters that seek analytical estimates based on a closed-form Markov–Bayes recursion, e.g., recursion from a Gaussian or Gaussian mixture (GM) prior to a Gaussian/GM posterior (termed ‘Gaussian conjugacy’ in this paper), form the backbone for a general time series filter design. Due to challenges arising from nonlinearity, multimodality (including target maneuver), intractable uncertainties (such as unknown inputs and/or non-Gaussian noises) and constraints (including circular quantities), etc., new theories, algorithms, and technologies have been developed continuously to maintain such a conjugacy, or to approximate it as close as possible. They had contributed in large part to the prospective developments of time series parametric filters in the last six decades. In this paper, we review the state of the art in distinctive categories and highlight some insights that may otherwise be easily overlooked. In particular, specific attention is paid to nonlinear systems with an informative observation, multimodal systems including Gaussian mixture posterior and maneuvers, and intractable unknown inputs and constraints, to fill some gaps in existing reviews and surveys. In addition, we provide some new thoughts on alternatives to the first-order Markov transition model and on filter evaluation with regard to computing complexity.

Clustering for filtering

Multi-sensor multi-object detection and estimation is solved by a clustering approach.Accommodate little prior information about targets, background and sensors.Neither sophisticated modeling nor unrealistic assumption is required.Outperform state-of-the-art filters in average multi-sensor cases. Advanced multi-sensor systems are expected to combat the challenges that arise in object recognition and state estimation in harsh environments with poor or even no prior information, while bringing new challenges mainly related to data fusion and computational burden. Unlike the prevailing Markov-Bayes framework that is the basis of a large variety of stochastic filters and the approximate, we propose a clustering-based methodology for multi-sensor multi-object detection and estimation (MODE), named clustering for filtering (C4F), which abandons unrealistic assumptions with respect to the objects, background and sensors. Rather, based on cluster analysis of the input multi-sensor data, the C4F approach needs no prior knowledge about the latent objects (whether quantity or dynamics), can handle time-varying uncertainties regarding the background and sensors such as noises, clutter and misdetection, and does so computationally fast. This offers an inherently robust and computationally efficient alternative to conventional Markov-Bayes filters for dealing with the scenario with little prior knowledge but rich observation data. Simulations based on representative scenarios of both complete and little prior information have demonstrated the superiority of our C4F approach.

Multi-target detection and estimation with the use of massive independent, identical sensors

This paper investigates the problem of using a large number of independent, identical sensors jointly for multi-object detection and estimation (MODE), namely massive sensor MODE. This is significantly different to the general target tracking using few sensors. The massive sensor data allows very accurate estimation in theory (but may instead go conversely in fact) but will also cause a heavy computational burden for the traditional filter-based tracker. Instead, we propose a clustering method to fuse massive sensor data in the same state space, which is shown to be able to filter clutter and to estimate states of the targets without the use of any traditional filter. This non-Bayesian solution as referred to massive sensor observation-only (O2) inference needs neither to assume the target/clutter model nor to know the system noises. Therefore it can handle challenging scenarios with few prior information and do so very fast computationally. Simulations with the use of massive homogeneous (independent identical distributed) sensors have demonstrated the validity and superiority of the proposed approach.

On generalized covariance intersection for distributed PHD filtering and a simple but better alternative

Some concerns are raised on the prevailing generalized covariance intersection (GCI) based Gaussian mixture probability hypothesis density (GM-PHD) fusion for distributed multiple target tracking under cluttered environments, which is both communicative and computation expensive, and generates a large amount of Gaussian components (GCs) of little physical significance. The problems become more serious when targets are closely distributed and/or when clutter is heavy. To avoid these problems and to save communication and computation, we advocate to only share the sufficiently strong-weighted GCs between neighboring sensors. The shared significant GCs are simply merged based on their spatial proximity, which resembles a type of multisensor signal superposition and will enhance the signal-noise-ratio (SNR) since strong GCs are more likely to be a “target signal” than a weak one, thereby facilitating less likely false alarms and a more accurate estimation. In parallel to the conservative GC sharing and merging, a standard averaging consensus is also sought on the cardinality distribution (a.k.a. the probability distribution of the target number) among sensors. Simulations have been provided to demonstrate the superiority and reliability of our approach with comparison to the benchmark GCI approach.

Convergence of Distributed Flooding and Its Application for Distributed Bayesian Filtering

Distributed flooding is a fundamental information sharing method to obtaining network consensus via peer-to-peer communication. However, a unified consensus-oriented formulation of the algorithm and its convergence performance are not explicitly available in the literature. To fill this void in this paper, set-theoretic flooding rules are defined by encapsulating the information of interest in finite sets (one set per node), namely distributed set-theoretic information flooding (DSIF). This leads to a new type of consensus called “collecting consensus,” which aims to ensure that all nodes get the same information. Convergence and optimality analyses are provided based on a consistent measure of the degree of consensus of the network. Compared with the prevailing averaging consensus, the proposed DSIF protocol benefits from avoiding repeated use of any information and offering the highest converging efficiency for network consensus while being exposed to increasing node-storage requirements against communication iterations and higher communication load. The protocol has been advocated for distributed nonlinear Bayesian filtering, where each node operates a separate particle filter, and the collecting consensus is sought on the sensor data alone or jointly with intermediate local filter estimates. Simulations are provided to demonstrate the theoretical findings.