Gee Wah Ng

Sensor management – what, why and how

Abstract This paper presents a comprehensive discussion and systematic classification of the sensor management (SM) with respect to its roles, concepts, architectures and techniques. The SM is put in perspective with the discussion of the sensing system framework. The role of SM was presented based on the different level of its functionality. The architectures discussed in this paper include centralized, decentralized and hierarchical methods. SM techniques such as scheduling and decision-making techniques are discussed. A generic SM framework is also presented.

Multiobjective optimization of sensor network deployment by a genetic algorithm

Decision support tools for assisting the human mission planner in the deployment of sensor networks is an important component of sensor management. The optimal selection of the number and types of sensors available from a suite of sensors, and their optimal placement in the terrain, is typically a multiobjective optimization problem with objectives defined based on the mission and scenario. One of the key advantages in applying multiobjective genetic algorithms for solving such problems is their ability to find multiple Pareto optimal solutions in a single run which then allows for the mission planner to select a final optimal solution based on higher-level considerations. The aim in this work is an effective genetic algorithm implementation of such a decision support tool for the deployment of sensor networks.

A cognitive architecture for knowledge exploitation

A cognitive architecture specifies a computational infrastructure that defines the various regions and functions working as a whole to produce human-like intelligence [Newell, 1990]. It also defines the main connectivity and information flow between the various regions and functions. These functions and the connectivity between them in turn facilitate and provide implementation specifications for a variety of algorithms. Drawing inspirations from computational science, neuroscience, psychology and biology, a top-level cognitive architecture which models the information processing in human brain is developed. Four key design principles [Ng, 2009] inspired by the brain, namely the hierarchical structure, distributed memory, parallelism in information flow and pathways, are incorporated into the architecture. A prototype cognitive architecture is developed and it is able to bring to bear different types of knowledge to solve a problem. It has been applied to object recognition in images. The cognitive architecture is able to exploit bottom-up perceptual information, top-down contextual knowledge and visual feedback in a way similar to how human utilizes different knowledge to recognize objects in images.

RF Emitter Geolocation using Amplitude Comparison with Auto-Calibrated Relative Antenna Gains

We consider an RF emitter localization problem that consists of a stationary RF emitter as a target and a moving RF receiver as an observer. An antenna system with unknown bias is deployed in the RF receiver to find the target direction of arrival (DOA) or bearing. This problem is conventionally solved by two separate stages, namely, to estimate target bearings through a relative amplitude direction finding (DF) algorithm, and to find the emitter location from the estimated bearings. The drawback of such a structure is that it is difficult to identify antenna bias through the DF algorithm alone, and the bearing error caused by antenna bias can significantly decrease the localization accuracy, especially when a target is located in longer range. To overcome this drawback, we merge the two stages together to formulate the localization problem as a discrete dynamic estimation problem. Thus, the target location and the antenna bias can be estimated simultaneously. Two algorithms are developed to cope with the merged system. The first algorithm, which is named as A-UKF, uses an unscented Kalman filter (UKF) to estimate the target location and the antenna bias in an augmented state. The second algorithm called MM-UKF further improves the A-UKF by introducing a multiple model (MM) approach to overcome the problem caused by an inaccurate initial state. The inaccurate initial state can seriously affect the performance of subsequent nonlinear estimation. The results show that both proposed algorithms are obviously superior to the conventional two-stage localization algorithm, and the MM-UKF algorithm outperforms the A-UKF.

UGHF for acoustic tracking with state-dependent propagation delay

Acoustic tracking with propagation delay is a challenging problem for the following reasons: It is difficult to perform an accurate state prediction, as the time interval between the current state and the previous state is varying and unknown due to the propagation delay; and the target time (signal emission time) needs to be estimated in addition to the target position and velocity. With the state augmented with the target time, the state transition cannot be described by the commonly used explicit Gauss-Markov model. In this paper, we propose a new approach to solving this difficult problem by using the Gauss-Helmert model for the state transition, which consists of an implicit equation between two consecutive states. An unscented Gauss-Helmert filter is then developed based on this formulation. The new approach is applied to the bearings-only tracking problem with state-dependent propagation delay. Simulation tests are conducted to demonstrate the performance of the unscented Gauss-Helmert filter, which is shown to outperform other approaches in terms of estimation accuracy, especially when a target is moving at high speed.

Classification for overlapping classes using optimized overlapping region detection and soft decision

In many real applications such as target detection and classification, there exist severe overlaps between different classes due to various reasons. Traditional classifiers with crisp decision often produce high rates of mis-classifications for patterns in overlapping regions. In this paper, we propose to use soft decision strategy with an optimized overlapping region detection to address the overlapping class problem. In contrast to crisp decision that assigns a single label to a pattern, the soft decision strategy provides multiple decision options to system operators for further analysis, which is believed to be better than producing a wrong classification. The effectiveness of the proposed method has been tested on both artificial and real-world problems.

Application of intent inference for surveillance and conformance monitoring to aid human cognition

Intent inference involves analyzing the actions and activities of a target of interest to deduce its purpose. In an environment cluttered with many targets, loaded with information, and under stress, the human may not be able to perform well Hence a cognitive aid that could derive possible intent inference and monitor the target may help augment human cognition and assist critical human decision making. This paper reports research on two applications: determining the likelihood of weapon delivery by an attack aircraft under military surveillance and conformance monitoring in air traffic control systems. The proposed solution is based on flight profile analysis. Simulation process comprises Interacting Multiple Model-based state estimation and Mamdani-type fuzzy inference to deduce possibilities of weapon delivery and of nonconforming aircraft behavior respectively. Results verify that the method is feasible and provides timely inference that will aid human cognition and hence assist decision making.

Target tracking in wireless sensor networks using particle filter with quantized innovations

Due to the bandwidth constraint of wireless sensor networks, there can be physical limitations in the communication links from sensors back to fusion center, or between sensors. In such cases, local data quantization/compression is not only a necessity, but also an integral part of the design of the sensor networks. In this paper, a target tracking approach using particle filter with quantized innovations in wireless sensor networks is proposed. The posterior Cramer-Rao lower bound for quantized innovation information received by fusion center is also given. The simulation results show the good performance of our proposed tracking approach. With a moderate small number of particles sampled at each step, we found that the tracking performance of particle filter is much better than the EKF, especially when the emitted power of each sensor is small.

Combining IMM Method with Particle filters for 3D maneuvering target tracking

The interacting multiple model (IMM) algorithm is a widely accepted state estimation scheme for solving maneuvering target tracking problems, which are generally nonlinear. During the IMM filtering process, serious errors can arise when a Gaussian mixture of posterior probability density functions is approximated by a single Gaussian. Particle filters (PFs) are effective in dealing with nonlinearity and non-Gaussianity. This work considers an IMM algorithm that includes a constant velocity model, a constant acceleration model and a 3D turning rate (3DTR) model for tracking three-dimensional (3D) target motion, using various combinations of nonlinear filters. In existing literature on combining IMM and particle filtering techniques to tackle difficult target maneuvers, a PF is usually used in every model In comparison, simulation results show that by using a computationally economical PF in the 3DTR model and Kalman filters in the remaining models, superior performance can be achieved with significant reduction in computational costs.

Adaptive Fuzzy Rule-Based Classification System Integrating Both Expert Knowledge and Data

This paper presents an adaptive fuzzy rule-based classification system using a new hybrid modeling method that integrates both expert knowledge and new knowledge learnt from data. Inspired by human learning, the membership functions of fuzzy rules are optimized based on a hybrid error function that combines errors caused by the class predefined by expert knowledge and nearby historical data. The weights of the two errors can be adjusted by a conservative parameter. Experimental results show that our method significantly reduces classification ambiguity in 9 datasets.