Chee-Yee Chong

Distributed Tracking in Distributed Sensor Networks

A distributed sensor network (DSN) consists of a set of processing nodes collecting data from sensors. When the nodes communicate, each node fuses the information received from other nodes with the local information to update its estimate on the state of the world. Previous papers have dealt with the theoretic algorithms for tracking multiple targets using a DSN. This paper considers the processing functions performed by each DSN node and presents some simulation results on tracking vehicles moving over a road network.

Track association and track fusion with nondeterministic target dynamics

Representative track fusion algorithms and track association metrics are quantitatively compared using a simple linear-Gaussian-Poisson model, under various degrees of nondeterministicity of the target dynamics, i.e., process noises, and of the initial condition uncertainty. Track fusion algorithms are compared using an analytical method, while track association metrics are evaluated by Monte Carlo simulations.

Distributed tracking in wireless ad hoc sensor networks

Abstract : Target tracking is an important application for wireless ad hoc sensor networks. Because of the energy and communication constraints imposed by the size of the sensors, the processing has to be distributed over the sensor nodes. This paper discusses issues associated with distributed multiple target tracking for ad hoc sensor networks and examines the applicability of tracking algorithms developed for traditional networks of large sensors. when data association is not an issue, the standard pre- predict/update structure in single target tracking can be used to assign individual tracks to the sensor nodes based on their locations. Track ownership will have to be carefully migrated, using for example information driven sensor tasking, to minimize the need for communication when targets move. when data association is needed in tracking multiple interacting targets, clusters of tracks should be assigned to groups of collaborating nodes. Some recent examples of this type of distributed processing are given. Keywords: Wireless ad hoc sensor networks, multiple target tracking, distributed tracking

Information Fusion in Distributed Sensor Networks

A distributed sensor network (DSN) consists of a set of processing nodes collecting data from sensors. When the nodes communicate, each node fuses the information received from other nodes with the local information to obtain an updated situation assessment. This paper presents fusion results based on a multiple hypothesis approach. The two important problems of hypothesis formation and evaluation are discussed.

MAP track fusion performance evaluation

The purpose of this paper is to develop a quantifiable performance evaluation method for MAP (Maximum A Posterior) track fusion algorithm. The goal is to provide analytical fusion performance without extensive Monte Carlo simulations. The idea is to develop methodologies for steady state fusion performance. Several fusion algorithms such as simple convex combination, cross-covariance combination (CC), information matrix (IM), and MAP fusion have been studied and several performance evaluation methods have been proposed. But most of them are not based on the steady state of an actual dynamic system. This paper conducts similar analysis for MAP fusion algorithm. It has been shown that the MAP or Best-Linear Unbiased Estimate (BLUE) fusion formula provides the best linear minimum mean squared estimates (LMMSE) given local estimates under the linear Gaussian assumption in a static situation (i.e., single iteration). However, in a dynamic situation, recursive fusion iterations are needed and the impact on the performance is not obvious. This paper proposes a systematic analytical procedure to evaluate the performance of such algorithm under two different communication strategies. Specifically, hierarchical fusion with and without feedback is considered. Theoretical curves for the steady state performance of the fusion algorithm with various communication patterns are given. They provide performance bounds for different operating conditions.

Generalized Murty's algorithm with application to multiple hypothesis tracking

This paper describes a generalization of Murtys algorithm generating ranked solutions for classical assignment problems. The generalization extends the domain to a general class of zero-one integer linear programming problems that can be used to solve multi-frame data association problems for track-oriented multiple hypothesis tracking (MHT). The generalized Murtys algorithm mostly follows the steps of Murtys ranking algorithm for assignment problems. It was implemented in a hybrid data fusion engine, called All-Source Track and Identity Fusion (ATIF), to provide a k- best multiple-frame association hypothesis selection capability, which is used for output ambiguity assessment, hypothesis space pruning, and multi-modal track outputs.

Evaluating a multiple-hypothesis multitarget tracking algorithm

The performance evaluation of multiple-hypothesis, multitarget tracking algorithm is presented. We are primarily interested in target-detection/track-initiation capabilities as measures of performance. Through Monte Carlo simulations, a multiple-hypothesis tracking algorithm was evaluated in terms of 1) probability of establishing a track from target returns and 2) false track density. A radar was chosen as the sensor, and a general multiple-hypothesis, multitarget tracking algorithm was used in the Monte Carlo simulations. The simulation results predict the probability of establishing a track from returns of a target as well as the false track density per scan volume per unit time. The effects of the target radar cross section and the radar power, measured through the mean signal-to-noise ratio (SNR) were studied, as were the effects of detection threshold and track quality threshold. Computational requirements were also investigated. >

Adaptive distributed estimation

This paper considers the distributed estimation problem when the communication pattern among the agents is not known a priori. We assume a network of estimation agents which receive measurements from the environment. The estimation agents communicate with each other using schedules which may not be fixed a priori. The objective of each agent is to generate the best estimate at any moment based on the local data and information received from other agents. Because the network configuration and/or communication pattern may change, the fusion algorithm of each agent cannot be specified a priori but has to adapt to the structure of the network. A fusion algorithm which can adapt is presented in this paper. The algorithm is based on the partial information graph available to that agent and makes use of the estimates as well as the history of communication.

Tracking Aircraft by Acoustic Sensors; Multiple Hypothesis Approach Applied to Possibly Unresolved Measurements

Poor resolution of acoustic sensors frequently lead to merged measurements for closely spaced targets. This paper considers tracking low-altitude aircraft by a network of acoustic sensors. Merging measurement outputs from sensors are probabilistically analysed. A multiple hypothesis approach is then used to derive an algorithm for tracking the targets. The likelihood functions used in hypothesis evaluation are derived assuming two-way merging and a simulated example is used to illustrate the algorithm.

Performance prediction of feature-aided track-to-track association

This paper describes analytic and semianalytic methods for predicting performance of track-to-track association, in terms of correct association probability, by an optimal assignment algorithm. The focus of this paper is to quantify how much feature or attribute information may improve association performance over the standard kinematic-only track-to-track association. Our goal is to obtain an analytical formula to predict the association performance as a function of a set of key parameters that quantify the quality of feature information. The result extends our previous development of an exponential law for predicting association performance, by including the effects of the additional generally non-Gaussian feature or attribute information.This paper is concerned with analytical and semi-analytical methods for predicting performance of track-to-track association, in terms of probability of each track being correctly associated with the track that shares the same origin, when association is performed by an optimal assignment algorithm. The focus of this paper is to quantify how much feature or attribute information can be expected to improve association performance over the usual track-to-track association using only kinematic or geolocational information. Our goal is to obtain a simple formula to predict the performance as a function of a set of key parameters that quantify the quality of feature information. The result extends the existing framework, which we may call the exponential law to predict association performance, to include the effects of the feature information.