In this paper, a new state estimation algorithm for estimating the states of targets that are separable into
linear and nonlinear subsets with non-Gaussian observation noise distributed according to a mixture of Gaussian
functions is proposed. The approach involves modeling the collection of targets and measurements as random
finite sets and applying a new Rao-Blackwellised Approximate Conditional Mean Probability Hypothesis Density
(RB-ACM-PHD) recursion to propagate the posterior density. The RB-ACM-PHD filter jointly estimates the
time-varying number of targets and the observation sets in the presence of data association uncertainty, detection
uncertainty, noise and false alarms. The proposed algorithm approximates a mixture Gaussian distribution with a
moment-matched Gaussian in the weight update phase of the filtering recursion. A two dimensional maneuvering
target tracking example is used to evaluate the merits of the proposed algorithm. The RB-ACM-PHD filter
results in a significant reduction in computation time while maintaining filter accuracies similar to the standard
sequential Monte Carlo PHD implementation.
The Interacting Multiple Model (IMM) estimator has been proven to be effective in tracking agile targets.
Smoothing or retrodiction, which uses measurements beyond the current estimation time, provides better estimates
of target states. Various methods have been proposed for multiple model smoothing in the literature.
In this paper, a new smoothing method, which involves forward filtering followed by backward smoothing while
maintaining the fundamental spirit of the IMM, is proposed. The forward filtering is performed using the standard
IMM recursion, while the backward smoothing is performed using a novel interacting smoothing recursion.
This backward recursion mimics the IMM estimator in the backward direction, where each mode conditioned
smoother uses standard Kalman smoothing recursion. Resulting algorithm provides improved but delayed estimates
of target states. Simulation studies are performed to demonstrate the improved performance with a
maneuvering target scenario. The comparison with existing methods confirms the improved smoothing accuracy.
This improvement results from avoiding the augmented state vector used by other algorithms. In addition, the
new technique to account for model switching in smoothing is a key in improving the performance.
This paper discusses a target tracking system that provides improved estimates of target states using target
orientation information in addition to standard kinematic measurements. The objective is to improve state
estimation of highly maneuverable targets with noisy kinematic measurements. One limiting factor in obtaining
accurate state estimates of highly maneuvering targets is the high level of uncertainty in velocity and acceleration.
The target orientation information is helpful in alleviating this problem to accurately determine the velocity and
acceleration components. However, there is no sensor that explicitly measures target orientation. In this paper,
the Observable Operator Model (OOM) is used together with multiple sensor information to estimate target
orientation measurement. This is done by processing the sensor feature measurements from different aspect
angles and the estimated target orientation measurement is used in conjunction with kinematic measurements
to conclusively estimate target states. Simulation results show that the incorporation of target orientation can
enhance the tracking performance in the presence of fast moving and/or maneuvering targets. In addition, the
Posterior Cramer-Rao lower bound (PCRLB) that quantifies the achievable performance is derived. It is shown
that the proposed estimator meets the PCRLB.
The joint target tracking and classification using target-to-sensor aspect-dependent Radar Cross Section (RCS)
and kinematic data for multistatic sonar network is presented in this paper. The scattered signals measured from
different orientations of a target may vary due to aspect-dependant RCS. A complex target may contain several
dozen significant scattering centers and dozens of other less significant scatterers. Because of this multiplicity
of scatterers, the net RCS pattern exhibits high variation with aspect angle. Thus, radar cross sections from
multiple aspects of a target, which are obtained via multiple sensors, will help in accurately determining the target
class. By modeling the deterministic relationship that exits between RCS and target aspect, both the target class
information and the target orientation can be estimated. Kinematic data are also very helpful in determining the
target class as it describes the target motion pattern and its orientation. The proposed algorithm exploits the
inter-dependency of target state and the target class using aspect-dependent RCS and kinematic information in
order to improve both the state estimates and classification of each target. The simulation studies demonstrate
the merits of the proposed joint target tracking and classification algorithm based on aspect-dependant RCS and
The Observable Operator Model (OOM) approach have been proposed as a better alternative to the Hidden
Markov Model (HMM). However the basic modeling of OOMs assume that the data is generated by some discrete
state variable which can take on one of several values which is unreasonable for most classification problems.
Main limitation of existing OOM classification is that they require substantial training data, assumed to be similar
to the data on which the algorithm is tested. In many applications the target is observed from multiple
target-sensor orientations (or aspects), and the underlying feature information is highly aspect dependant and
continuous variable. The multi-aspect target classification method presented based on continuous-valued Observable
Operator Model (OOM), from which a full posterior distribution of a target class is inferred. It is possible
to extend a discrete OOM as a continuous-valued OOM using a membership function. Further, predefined set
of classes were used in training based joint target tracking and classification methods. These methods perform
poorly, when new target present in the surveillance region which is not in the available class-set. In order to
overcome this shortage, we propose an online training algorithm for OOM, which identifies new incoming target
classes and add them into the available class-set. As the number of target class increases with the online learning
procedure, there is a need for an adaptive class-set selection in order to reduce computational cost. An adaptive
class-set approach for joint target tracking and classification is formulated via hypotheses testing, which reduces
computation cost compared to calculating OOM likelihood for each target class.
Simulation results are given to demonstrates the merits of continuous-valued Observable Operator Method
(OOM) for target classification over discrete OOM, advantages of online training OOM and the efficiency of
class-set adaptation algorithm.
In this paper, a new joint target tracking and classification technique based on Observable Operator Models (OOM) is considered. The OOM approach, which has been proposed as a better alternative to the Hidden Markov Model (HMM), is used to model the stochastic process of target classification. These OOMs afford both mathematical simplicity and algorithmic efficiency compared to HMM. Conventional classification techniques use only the feature information from target signatures. The proposed OOM based classification technique incorporates the target-to-sensor orientation together with a sequence of feature information from multiple sensors. The target-to-sensor orientation evolves over time and the target aspect is important in determining the target classes. The multi-aspect classification is modeled using OOM to handle unknown target orientation. This algorithm exploits the inter-dependency of target state and the target class, which improves both the state estimates and classification of each target. Measurement ambiguity is present in both kinematic and feature measurement and therefore, the OOM based classifier is integrated into the multiframe data association framework that is used to resolve measurement origin uncertainties. This technique enables one to overcome ambiguity in feature measurements while improving track purity. A two dimensional example demonstrates the merits of the proposed OOM based joint target tracking and classification algorithm.
Particle filter based estimation is becoming more popular because it has the capability to effectively solve nonlinear and non-Gaussian estimation problems. However, the particle filter has high computational requirements and the problem becomes even more challenging in the case of multitarget tracking. In order to perform data association and estimation jointly, typically an augmented state vector of target dynamics is used. As the number of targets increases, the computation required for each particle increases exponentially. Thus, parallelization is a possibility in order to achieve the real time feasibility in large-scale multitarget tracking applications. In this paper, we present a real-time feasible scheduling algorithm that minimizes the total computation time for the bus connected heterogeneous primary-secondary architecture. This scheduler is capable of selecting the optimal number of processors from a large pool of secondary processors and mapping the particles among the selected processors. Furthermore, we propose a less communication intensive parallel implementation of the particle filter without sacrificing tracking accuracy using an efficient load balancing technique, in which optimal particle migration is ensured. In this paper, we present the mathematical formulations for scheduling the particles as well as for particle migration via load balancing. Simulation results show the tracking performance of our parallel particle filter and the speedup achieved using parallelization.
Proc. SPIE. 5809, Signal Processing, Sensor Fusion, and Target Recognition XIV
KEYWORDS: Particles, Particle filters, Detection and tracking algorithms, Electronic filtering, Nonlinear filtering, Filtering (signal processing), Monte Carlo methods, Target detection, Digital filtering, Complex systems
The particle filter is an effective technique for target tracking in the presence of nonlinear system model, nonlinear measurement model or non-Gaussian noise in the system and/or measurement processes. In this paper, we compare three particle filtering algorithms on a spawning ballistic target tracking scenario. One of the algorithms, the tagged particle filter (TPF), was recently developed by us. It uses separate sets of particles for separate tracks. However, data association to different tracks is interdependent. The other two algorithms implemented in this paper are the probability hypothesis density (PHD) algorithm and the joint multitarget probability density (JMPD). The PHD filter propagates the first order statistical moment of multitarget density using particles. While, the JMPD stacks the states of a number of targets to form a single particle that is representative of the whole system. Simulation results are presented to compare the performances of these algorithms.
The particle filter is an effective technique for target tracking in the presence of nonlinear system model, nonlinear measurement model or non-Gaussian noise in the system and/or measurement processes. However, the current particle filtering algorithms for multitarget tracking suffer from high computational requirements. In this paper, we present a new implementation of the particle filter, called the tagged particle filtering (TPF) algorithm, to handle multitarget
tracking problems in an efficient manner. The TPF uses a separate sets of particles for each track. Here, each particle is associated with the closest (in terms of likelihoods) measurement. The particles for a particular track may form separate groups in terms of the measurements associated with them and they evolve independently in groups till two or more groups of particles are separated by a distance large enough to be called separate tracks. A
decision is made as to which of the groups is to be retained. Since
this algorithm keeps a separate set of particles for each track, the state estimation for individual tracks does not require any additional computation. Also, this algorithm is association free and target class information can be added to the state for feature aided
tracking. Simulation results are obtained by applying this tracking filter to a spawning target scenario.