School of Computer Science and Software Engineering

Fusion of multi-sensor data enhances the performance of systems. Fusion at the score and decision levels has produced improved performance in multimodal biometrics and terrain classification. It is believed that fusion at the data and feature levels will outperform the score and decision level fusion techniques. However, limited research has been done in this area due to a number of challenges (e.g. incompatibility, high dimensionality). This thesis will focus on data and feature level fusion. It will produce unified representations for multimodal data for applications in biometrics and to off-road robot navigation (terrain classification, obstacle detection and landmark-based localization). Hybrid fusion (data/feature levels and local/global features) algorithms will also be developed for these applications. The performance of the unified representations will be tested using ROC curves and will be compared with the benchmark performance (score level).

Back to index

Mr Lucas Bradstreet

Enrolment Type: PhD Full-time

Research Title: Transformation of Multi-objective Optimisation Problems to Single Objective Problems via Hypervolume

Supervisor(s): Dr Luigi Barone and Dr Lyndon While

Research Group: Adaptive Systems

Synopsis :

Multi-objective evolutionary algorithms (MOEAs) are an important new field. MOEAs are used for optimisation problems where trade-offs between multiple objectives may be required and where trade-offs cannot be predicted in advance, where trade-offs change over time (e.g. depending on market conditions), or where "engineering judgement" is required to make the optimal design decision. Cases where MOEAs have been applied successfully to real world problems include Barone's rock crusher MOEA and network layout problems.

There are many issues encountered when using MOEAs, such as good methods for diversity maintenance. Given the many applications of MOEAs to many current problems and the immaturity of the field, it is a fertile field for research into techniques which allow them to

operate more efficiently and effectively.

Back to index

Mr Daniel Bond

Enrolment Type : PhD Full-time

Research Title : Epistemic Logic

Supervisors : A Prof Mark Reynolds

Research Group: Formal Methods

Synopsis :

Back to index

Mr Bobby Chu

Enrolment Type : PhD Full-time

Research Title : Design Method for Bushfire Sensor System

Supervisors : Professor George Milne and Joel Kelso

Research Group: Formal Methods

Synopsis :

Bobby is focused on exploring modelling methodologies with respect to Wireless Sensor Networks in wildfire scenarios. He is using an Interacting Automata approach to capture the spatial features of wireless radio communication including the effects of wildfire on communication. Through the formal description of this system, simulators can be developed to determine best usage of sensors in a several wildfire applications. These include early detection systems and real-time data gathering for more effective fire suppression, and post-fire data retrieval to aid refinement of fire behaviour models.

Back to index

Mr Andrew Czarn

Enrolment Type : Part-Time PhD Student

Research Title : Statistical Exploratory Analysis of Genetic Algorithms

Supervisors : Dr Cara MacNish and Associate-Professor Kaipillil Vijayan

Research Group: Adaptive Systems

Synopsis:

Adaptive algorithms such as GAs work by iteratively adapting members of a population of potential solutions. While the mechanics of each individual adaptation are quite straightforward, the way individual changes affect the success of the population as a whole is more difficult to determine. This is also true of the many parameters that are used to fine tune, or improve the success of, adaptive algorithms. Examples include population size, mutation and crossover rates, and so on. Values for these parameters are most commonly set through a process of trial and error, or based on recommendations from related problems in the literature, rather than through statistically sound analysis of their affects on GA performance. This thesis proposes a rigorous yet practical statistical methodology for assessing the impact of parameter settings. The methodology addresses issues of experimental design, blocking, power calculation and response curve analysis.

Back to index

Mr John Darrington

Enrolment type : PhD Full time

Research Title : Optimisation of Image Processing for Real Time Operations

Supervisor(s) : Amitava Datta & Livia Hool (Physiology)

Research Group: Vision and Visualisation

Synopsis:

John is investigating real time signal processing techniques. Initially, he was focusing on image processing applications. However, he discovered that the algorithms I'm investigating have applications in ECG detection and recognition. Potential applications also exist for other physiological signals including respiration and blood pressure, and anything that needs fast detection/classification of a single dimensional signal.

Back to index

Ms Majigsuren Enkhsaikhan

Enrolment type:          PhD Full time
Research Title:           Detection of Ontology changes in Documents
Supervisor:                Dr Wei Lui and A.Prof Mark Reynolds

Research Group: Adaptive Systems

Synopsis:

Back to index

Mr Amer Filipovic

Enrolment type: PhD Full-Time

Research Title: Evironment-aware Localised Routing in Ad-Hoc Wireless Networks

Supervisors: Associate Professor Amitava Datta, Dr Chris McDonald

Research Group: Mobile, Ad Hoc and Sensor Networks

Synopsis:

Ad-Hoc wireless networks are the next generation improvised networks for devices that are on the move or in a location where networking infrastructure is not feasible. Amer is working on adaptive routing behaviour based on local information. This involves collecting information about the local neighbourhood and containing the information spread to the areas where it is needed the most. The resulting hybrid routing protocol of this research should significantly reduce the information overhead and improve the scalability of ad-hoc networks.

Back to index

Miss Elizabeth Fox

Enrolment Type : PhD Full-time

Research Title : Call-independent individual recognition in birds

Supervisors : A/Professor Mohammed Bennamoun, Prof. Dale Roberts and

Dr Alan Burbridge (Animal Biology)

Research Group: Vision and Visualisation

Synopsis :

Elizabeth aims to develop a new method of acoustically identifying individual birds. So far all work on acoustic identification has been done using a) single song types or limited song repertoires and b) relatively simple methods e.g. cross-correlation, eyeball comparisons of spectrograms, or discriminant function analysis. These techniques are limited in the extent of how finely they can dissect the signal and in their ability to deal with any complexity in the vocal repertoire. Developing a method of voice recognition will enable an individual to be identified regardless of the song type made. The ability to do this is likely to revolutionise individual recognition in birds, and other animals, and will be of particular value for work on threatened species which cannot be captured or marked by conventional methods.

Back to index

Mr Shih Ching Fu

Enrolment type: PhD Full-time

Research Title: Structure from Motion using Differential Invariants of Optical Flow

Supervisor: Dr Peter Kovesi

Research Group: Vision and Visualisation

Synopsis:

The change in the shape of objects caused by the relative motion between an object and observer can be decomposed into divergence, curl, and deformation components. These three components are related to 3D scene structure and ego-motion and determined from their affect on scene geometry. Through these relationships, these three quantities can be used to derive information about surface orientations and time-to-contact; two quantities which are useful in the Structure-from-Motion problem. My research examines the extraction of these quantities in the face of noisy optical flow using spatial filtering and contour moments.

Back to index

Mr Christopher Gunn

Enrolment Type : PhD Part-Time

Research Title : Using Haptics in a Networked Immersive 3D Environment

Supervisor : Professor Amitava Datta

Research Group: Vision and Visualisation

Synopsis :

The thesis examines the utility that haptics (or force feedback) provides a computer user. It does this by following a series of publications written by the author on various aspects of the subject and adds to these the results of some, as-yet unpublished, experiments. The thesis describes several immersive, 3D applications and prototypes that were developed by the author and his colleagues in the CSIRO Virtual Environments Laboratory between 1999 and 2006. The work shows that haptic feedback can successfully be integrated into artistic, planning and teaching environments and that in those cases it can enhance and the user’s perception of the virtual environment being depicted. It particularly focuses on the networking of these haptic applications - a topic which is covered by several of the papers as well as analysed more closely in the text. The author has been able to create collaborative haptic applications that run successfully over much larger distances than were previously thought possible, overcoming some of the problems introduced by the inherent latency of the network.

Back to index

Mr Tarek Hassan

Enrolment type : PhD Full time

Research Title : Computer vision engineering for detecting unexpected behaviour in video sequences

Supervisor(s) : A.Prof Mark Reynolds, Dr Tim French and A.Prof Mohammed Bennamoun

Research Group: Vision and Visualisation and Logic

Synopsis :

Mr Bob Hastings

Enrolment type : PhD Full time

Research Title : A Statistical Investigation of Fingerprint Patterns

Supervisor(s) : Dr Peter Kovesi

Research Group: Vision and Visualisation

Synopsis :

This project is concerned with the statistical uniqueness of fingerprints with regard to the features that can be used to classify them and to perform identification. We aim to gain an understanding of the information content, and the number of degrees of freedom, in fingerprint data, and ultimately to define criteria that will allow an individual to be identified to a defined confidence level using fingerprint information.

Back to index

Miss Sarah Hatton

Enrolment type : PhD Full time

Research Title : Software Requirements Prioritisation and the Management of Interactions.

Supervisor(s) : A/Prof Mark Reynolds

Research Group: Software Engineering

Synopsis :

Improve the efficiency and value of the requirements engineering process, through the refinement of the techniques of requirements prioritization, modelling of interactions and assignment of value.

Back to index

Mr Syed Mohammed Shamsul Islam

Enrolment Type : PhD Full time

Research Title : Unified Representation of Multimodal Biometrics for Robust Authentication and Identification.

Supervisor(s) : A/Prof. M. Bennamoun and Prof. Robyn Owens

Research Group: Vision and Visualisation

Synopsis:

Multimodal biometrics is a comparatively new research area where multiple physiological or behavioural characteristics of a user are taken into consideration for identification and verification purposes. Such combination considerably minimizes limitations of the individual biometrics. Approaches proposed so far are mostly fusing multiple biometrics at the score or decision level. To improve accuracy, we plan to make unified tensor-based 3D representation of prominent biometrics, such as face, ear, hand etc with their various modalities and fuse them at the data or feature extraction level. We will also develop appropriate database and testing methodology to asses the performance of the proposed system.

Back to index

Mr Keith Johnson

Enrolment type:          PhD Full time
Research Title:           TBA
Supervisor:                Dr Cara MacNish

Research Group: Adaptive Systems

Synopsis:

Back to index

Miss Angel Kennedy

Enrolment type:          PhD Full time
Research Title:           Enabling Higher Level Functions in Neural Networks Using Directed Processing
Supervisor:                Dr Cara MacNish

Research Group: Adaptive Systems

Synopsis:
Recent artificial neural network (ANN) models of working memory have demonstrated superior task switching and generalisation abilities over other ANNs. The main thing these models have in common is the ability to use activation memory to direct processing in other areas of the network relying on weight based memory. In many cases however use of directed processing is embedded within a more complex model and no efforts have yet been made to isolate the impact of directed processing. The aim of this research is to perform an investigation of the impact of directed processing on ANN performance. In addition we would like to investigate its potential to enable higher level processes such as look-ahead to be performed by ANNs.

Back to index

Mr Jonathan Knispel

Enrolment Type : PhD Full-Time

Research Title : Cooperating Modular Applications in 3D Virtual Environments

Supervisor(s) : A/Professor Amitava Datta, Assoc. Prof. Richard Thomas

Research Group: Vision and Visualisation

Synopsis :

Providing a higher-level framework for multiple 2D and 3D applications to cooperate in a shared 3D virtual environment simplifies application engineering and provides a richer, more flexible workspace for end-users. The 3D desktop workspace under development runs on commodity PC hardware and is designed to be simple for new users to learn, while providing a more flexible environment than existing 2D approaches. User trials will compare current 2D and 3D user interface techniques with new techniques developed for this environment.

Back to index

Miss Winnie Louis Lee

Enrolment type : PhD Full time

Research title : Network Management in Wireless Sensor Networks.

Supervisor(s) : Assoc. Prof. Amitava Datta, Dr Rachel Cardell-Oliver, Dr David Glance

Research Group: Mobile, Ad Hoc and Sensor Networks

Synopsis :

The use of wireless sensor networks for gathering environmental and safety-critical data in real time is increasing at a rapid rate. Some of the main criteria in designing sensor network architectures are energy-efficiency, self-management and self-healing. However, most protocols for data gathering and routing in sensor networks implicitly assume a regular rate of data gathering by individual nodes. While this is sufficient for sensing parameters that change slowly over time, individual nodes in a small part of a network may need to increase the rate of data gathering considerably for reporting important data in real-time. For example, in structural monitoring applications, sensor nodes are deployed to monitor vibration (e.g., wind and earthquakes) that could damage the structure of a building. It is critical for sensor nodes to send their data more often to the base station when they detect event triggers such as sensor readings changing rapidly or exceeding user-specified thresholds. Winnie’s research focuses on developing a flexible, reactive and energy efficient scheduled protocols for wireless sensor networks. Her protocols cover Medium Access Control, routing and application (management) layer issues. She has developed a novel fault-tolerant TDMA protocol for transferring time slots from one part of the network to another part to support non-uniform and reactive sensing in different parts of a network.

Back to index

Mr Nicholas Lowe

Enrolment Type : PhD Full-Time

Research Title : A Paradigm and Algorithms for Real-time Rendering of Complex Portals

Supervisor(s) : Assoc. Prof. Amitava Datta

Research Group: Vision and Visualisation

Synopsis :

Portal-based rendering is a common technique for increasing graphics rendering speed in real-time applications. It has recently been adapted for use in novel scene composition. This adaption is limited because the portal model and available rendering algorithms are severely constrained.

I have developed a new conceptual portal model that delineates between the intrinsic nature of portals and ancillary constraints. I have also developed algorithms to support rendering of the more complex portals that exist within this new model. These algorithms are designed to

complement the trend towards higher detail and greater programmability in real-time graphics.

Back to index

Mr Navid Mavaddat

Enrolment type: PhD Part-time

Title of Thesis: Single View Camera Based Text Recognition

Supervisor(s): Prof Robyn Owens, Dr Peter Kovesi

Research Group: Vision and Visualisation

Synopsis:

Navid’s thesis addresses the problem of recognising text placed within a 3D scene. Techniques for the recognition of a plain text in 2D planes are well established and documented, but the problem of text recognition in a 3D environment in the presence of other objects is of a different degree of complexity. For this situation, in addition to the transformation of the text from perspective projection to the plane of the text, the text plane and boundaries of the region where the text is to be found have to be identified and isolated from the rest of the image space. His work focuses on detection of the perspective distortion, the calculation of the correction transformation and finally the rectification of the image utilising the correction transformation. Both text written on planar and curved surfaces are considered. The application of this work includes automated reading for the

visually impaired, translation for travellers and text recognition for robots.

Back to index

Mr John C. McCabe-Dansted

Enrolment type : PhD Full Time

Research Title : Applications of modal logic to Computer Security.

Supervisor(s) : A/Prof Mark Reynolds

Research Group: Formal Methods

Synopsis :

John has just started his PhD. He am currently looking at proofs of completeness for modal logics, his supervisor believes this is necessary for effective use of modal logics. He has briefly looked at pre-existing proofs of security for cryptographic key-exchange; existing modal logic prove construction software; Typed Assembly Language, a mark-up for assembly language programs that proves that the assembly language program e.g. memory safe. He is also presenting a paper regarding the approximability of "Dodgson's rule", a voting procedure for which determining whether a Candidate is a winner is known to be an NP-hard problem.

Back to index

Mr Thomas Daniel Midgley

Enrolment type: PhD Part-time PhD

Title of Thesis: Dialogue grammar induction

Supervisor(s): Cara MacNish and Shelly Harrison

Research Group: Adaptive Systems

Synopsis:

Given an utterance in a dialogue, what should happen next? The answer will depend on who the next speaker is, what has already happened in the dialogue, and the intentions of the speakers. This thesis addresses the problem of how to induce a grammar for dialogue from a string of utterances in a dialogue corpus. This information would be useful in the construction of a dialogue manager for a natural language system, or in the dialogue act tagging task.

Adjacency pairs (e.g. question/answer) have a long history in the study of grammar, but most dialogue research has not used this information optimally. Dialogue is full of hesitation noises and disfluencies that interrupt the flow of conversation. We use statistical methods in combination with traditional grammar induction techniques to explain patterns of dialogue acts, reflect expectations in real dialogue, and predict future speech acts.

Back to index

Mr Babak Pazand

Enrolment Type: PhD Full-time

Research Title: Location-free Node Scheduling in Wireless Sensor Networks

Supervisor(s): A/Prof Amitava Datta, Dr. Rachel Cardell-Oliver
Research Group: Mobile, Ad Hoc and Sensor Networks

Synopsis: Node scheduling is the process of deciding the off-duty and on-duty

cycle for each sensor node. It determines the eligibility of a node to be active or inactive. At each time slice, selected working nodes perform the sensing tasks on behalf of other redundant nodes. There are two types of node scheduling, location aware and location free. Location dependant solutions rely on GPS devices which impose too much energy consumption and increase the cost of deployment. Also, current location free schemes suffer from some drawbacks such as too much packet and energy overhead. Babak’s research project has focused on devising a new location independent node scheduling scheme for wireless sensor networks. This solution is based on Minimum Dominating Sets, a graph theoretic formulation. Nodes discover their neighbours by a simple mechanism of exchanging control packets. Then, the base station builds the graph of network and determines a collection of minimum dominating sets of this graph. At each round one set is responsible to cover the network. Babak will investigate some forms of novel localization techniques to improve the coverage ratio.

Back to index

Mr Valance Phua

Enrolment Type : PhD Full-time

Research Title : A Framework for Wireless Sensor Networks in Manufacturing Environments

Supervisor(s) : A/Prof Amitava Datta and Dr Rachel Cardell-Oliver
Research Group: Mobile, Ad Hoc and Sensor Networks

Synopsis :

Valance’s research is focused on the efficient use of wireless sensor networks in a factory environment. Since the operating conditions in factories poses severe constraints to radio frequency communication, existing wireless sensor network implementations will not work well under these conditions because they do not consider several factors in the communication channel, such as multipath propagation and factory noise, which both lead to signal fading. As such, the research involves building a framework for wireless sensor networks that takes into account the undesirable properties of the communication channel in a typical factory environment while optimizing performance metrics such as energy-efficiency, network performance, and self-organization. This includes building a realistic simulation environment of the manufacturing plant, devising suitable Medium Access Control (MAC) and routing protocols specifically designed for industrial applications, and integrating network management protocols to improve network performance.

Back to index

Mr Anthony Prior

Enrolment type: PhD Full-time

Title: 6 Degrees-of-freedom Collaborative Haptic Virtual Sculpting

Supervisor(s): Dr Karen Haines and Dr Amitava Datta

Research Group: Vision and Visualisation

Synopsis:

Virtual sculpting allows users to created 3-dimensional artworks and prototypes by using a 3D input device to control a sculpting tool. While existing methods allow a range of sculpting effects, they are limited to simplistic polyhedral tool shapes (usually spherical or point-based) or complex voxel-based tools that can only be applied at fixed orientations. The research involves is the development of a model for real-time voxel-based virtual sculpting that performs on-the-fly voxelization of a polyhedral tool to determine the region of the sculpture to modify. This allows the tool to take on a variety of shapes ranging from simple to complex polyhedra and also allows the tool to be applied to the sculpture at any orientation. The model provides 6-degrees-of-freedom haptic feedback allowing users to feel what they are sculpting. The scene is visualized using a localized Marching Cues algorithm. As an extension to this work, Anthony is also developing a collaborative version of the sculpting model, allowing multiple users at remote locations to work on single sculpture simultaneously.

Back to index

Miss Lanny Sitanayah
Enrolment type :         MSc Full Time
Research Title :          Counting and tracking moving objects in wireless sensor networks without location information
Supervisor(s) :           A/Prof Amitava Datta and Dr Rachel Cardell-Oliver
Research Group:       Mobile, Ad Hoc and Sensor Networks

Synopsis :

Wireless Sensor Networks (WSN) is a developing research area which has lots of advantages for military, environmental, health, home and some commercial applications in detecting and monitoring variety of conditions without presence of human as operator directly. Moreover, one of the tasks of WSN which is necessary for military and habitat monitoring applications is tracking moving objects. This task is simpler if the exact location of each sensor, which is deployed in a monitoring area, is known. Unfortunately, the sensors are distributed randomly and building a WSN with special location hardware like GPS embedded in the sensors is extremely expensive compares with the sensors themselves. Therefore, this project is proposed to solve the counting and tracking of moving objects in WSN without knowing the location information of the sensors.

Back to index

Mr James Strauss

Enrolment Type: PhD Full-time

Title: View-Dependent Polygonal Mesh Refinement

Supervisor: Assoc. Prof. Amitava Datta

Research Group: Vision and Visualisation

Synopsis:

Developing view-dependent polygonal mesh refinement techniques for level-of-detail and non-photorealistic rendering. Much of the work involves designing data structures that store polygonal meshes in such a way that mesh resolution may be changed dynamically. In other words, areas of the mesh can be made finely or coarse detailed on-the-fly. The research examines the usefulness of these data structures in other graphics problems such as ray-tracing and scene shadowing.

Back to index

Ms Jingbo Sun

Enrolment type : PhD Full time

Research Title : A Framework for Building Autonomic Environmental Sensor Networks

Supervisor(s) : Rachel Cardell-Oliver
Research Group: Mobile, Ad Hoc and Sensor Networks

Synopsis :

As analysing the data from field test, we know that the performance of wireless sensor network in field test is unpredictable and unreliable. Jingbo is simulating some related routing protocols: ARQ stop and wait, Streaming Communication Model and Opportunistic routing: ExOR protocols, using the simulators built in C# programming language, and predicting and analysing their performance according to different metrics. Jingbo is going to build a reliable protocol which could have better performance in field test based on the result from simulation. Then she will try to implement it in real sensor motes and test its performance in field.

Back to index

Mr Christopher Thorne

Enrolment Type : PhD Full-Time

Research Title : Minimizing Position Dependent Errors in Computation on an Non-uniform Discrete Field and its Application to Improving Quality and Scalability of Computer Simulation

Supervisor(s) : A/Professor A. Datta

Research Group: Vision and Visualisation

Synopsis :

This work offers a new origin centric approach, including design, techniques and process, to improving the accuracy and scalability of simulation computation. It is applicable particularly to computer graphics applications but will likely find use in other forms of computer simulation. The focus of the work is on improvements that derive from a better understanding and exploitation of the floating point space used in computer simulation. The results of applying the approach will be to minimise error throughout the simulation pipeline from input through to final graphical output, leading to better performance, quality and scalability of a wide range of applications.

Back to index

Miss Minh Tue Tran

Enrolment Type: PhD Full time

Research Title: Patchwork Texture Synthesis with Applications to Large-Scale

3D Models

Supervisors: A/Professor Amitava Datta & Dr Nick Spadaccini

Research Group: Vision and Visualisation

Synopsis:

Texture can be used in computer graphics to depict surface detail which does not necessarily occur in the surface geometry of the object. Texturing 3D models not only improves visual aesthetics but can potentially increase the number and speed at which models are generated in a scene. However, the texture size may not always fit the model and so, texture extension is needed. The process of imitating a texture into various sizes and dimensions is known as texture synthesis. The goal is to imitate the sample texture in such a way that sample and synthesised texture are perceived to be generated by the same source. Previous attempts have produced undesirable artefacts such as image seams, smudging, and noise which are not inherent from the sample, and may require extensive user intervention. This thesis will focus on a texture synthesis method which, after a pre-processing step, can stitch together texture patches quickly to be used on large scale models with minimal user intervention. To reduce the amount of undesirable artefacts present within the output texture, the method will use image edges (high intensity gradients) to maintain image structure and continuity.

Back to index

Mr Daniel Wedge

Enrolment type : PhD full time

Research Title : Video Sequence Synchronization

Supervisor(s) : Dr Du Huynh & Dr Peter Kovesi

Research Group: Vision and Visualisation

Synopsis

I am working on methods of synchronizing pairs of video sequences, i.e., given two videos of the same event recorded simultaneously by two camera angles, I am using motion cues such as a ball's trajectory to recover the ratio of frame rates of the two cameras (since they may record at different frame rates) and the time delay between the beginnings of the two sequences (since the cameras may start recording at different times).

Back to index

Mr Andrzej Welna

Enrolment Type : PhD Part-time (currently suspended)

Research Title : Algorithms for shape reconstruction

Supervisors : A/Professor Ryszard Kozera

Research Group: Vision and Visualisation

Synopsis

The project focuses on different aspects of shape reconstruction from single and multiple image(s). It includes deriving novel algorithms (e.g. based on finite difference methods) for explicitly solving an unknown shape.

The shape from shading problem consists in extracting an unknown surface from its image shading. Recently in an effort to obtain more rigorous uniqueness results combined with the derivation of an explicit formula for the shape recovery, a Photometric Stereo technique has been developed. The purpose of these project is to investigate the usage of the numerical algorithms for implementing the theoretical results of Photometric Stereo technique and single image recovery to reconstruct the unknown surface. The aim of the project is to develop, implement and evaluate fast, provably convergent and stable algorithms applicable for the shape reconstruction problem based on a single (multiple) image input.

The problem has significant impact both theoretically and practically to researchers, and computer software developers. Any successful research in shape from shading area has potential applications in artificial intelligence, medical and satellite image processing and analysis as well as in improving our understanding of human visual system. Various methods and techniques in implementing mathematical theoretical models describing the real processes through appropriate numerical methods into the computer systems, combined with a different programming techniques and system development strategies are used to implement theoretical results into a computer based environment.

Back to index

Mr Mark Wittkamp

Enrolment type : PhD Full Time

Research Title : The use of Evolutionary Algorithms and Opponent Modelling for Continuous in-Game Adaptation Involving Complex Planning Tasks in Computer Games with Competing Agents

Supervisor(s) : Dr Luigi Barone and Dr Phil Hingston (Edith Cowan)

Research Group: Adaptive Systems

Synopsis

My research lies in artificial intelligence in video games using evolutionary algorithms and opponent modeling. More specifically, the problem of complex planning tasks in video games in the presence of competing individuals (i.e. computer of human players with different or opposing goals). For many video games there exists no general optimal playing strategy. Instead, optimal strategies will largely depend on the strategies employed by the competing individuals; this is the realm of opponent modeling.

It is often infeasible to use evolutionary algorithms directly to produce sufficiently adaptive and functional artificial game opponents, especially for very complex planning where real time learning is desired. Offline learning may be necessary due to the large number of input cases that are typically required for evolutionary algorithms to yield desirable results.

One approach may be that model representations of the opponent(s) are constructed and used in an evolutionary algorithm to develop complex game playing plans offline while the game progresses and refinements to the model occur. Previous work shows that computational intelligence techniques have been used for adaptation in similar problems. This work investigates using them for continuous adaptation in a video game AI.

Back to index

Mr Jason Wong

Enrolment type : PhD Full Time

Research Title : Developing a robust, realistic, and real-time animation system for modelling soft plants.

Supervisor(s) : A/Prof Amitava Datta

Research Group: Vision and Visualisation

Synopsis :

To develop a suitable technique to render and animate soft bodied models, such as small plants, in real-time. Also, the project involves modelling the effects of the virtual environment on the plant model, such as wind, rain and gravity. These parameters are changeable at any time, so the animation of the plant must be robust as well as efficient to remain in the real-time domain.

Back to index

Mr TzuYen Wong

Enrolment type : PhD Full time

Research Title : Quality measures of Image morphing

Supervisor(s) : Dr Peter Kovesi, A/Prof Amitava Datta

Research Group: Vision and Visualisation

Synopsis :

My research is a mixture of computer vision and computer graphics. I study the epipolar geometry of how 3D world is projected to 2D images and the reversed relationships. I also study the various techniques of image morphing, ie. smooth transformation from one image to another. I devise techniques to quantitatively measure the merits of morphing techniques and morphing sequences. I hope those measures will reveal more insights about image morphing leading to better morphing techniques and more amazing computer graphics.

Back to index

Mr Wilson Yik-Sen Wong

Enrolment type : PhD Full-time

Research Title : Ontology Maintenance in a Knowledge-based Question

Answering Environment

Supervisor(s) : Dr Wei Liu and A/Prof Mohammed Bennamoun

Research Group: Vision and Visualisation

Synopsis :

Ontology is essential to knowledge-based question answering systems for assisting in answer discovery and advanced reasoning. Despite the importance of ontology, the process of construction and maintenance remains manual, leading to poor extensiveness in existing knowledge-based question answering systems. To aggravate the situation, the idea to automate the process of constructing and maintaining ontology is faced with many problems. This research offers a quid pro quo solution to the poor extensiveness in knowledge-based question answering and the problems related to automatic maintenance of ontology.

Back to index

Mr Joseph Ziegler

Enrolment type : PhD Full time

Research title : Data Management in Wireless Sensor Networks

Supervisor(s) : Dr Rachel Cardell-Oliver

Research Group: Mobile, Ad Hoc and Sensor Networks

Synopsis :

Back to index

CSSE PhD and MSc Graduates

Dr Simon Collings PhD 2007

Frontier Points: Some theorems and methods for computer vision

Supervisors : A/Professor Ryszard Kozera and J Noakes

Research Group: Vision and Visualisation

Synopsis:

In stereo vision, a frontier point F belonging to an observed surface has the property that the tangent plane to the surface at F coincides with the epipolar plane at F. Frontier points can be estimated from a stereo pair without solving the classical stereo correspondence problem and one then knows the orientation of the surface at these points.

In the case where a large number of frontier points is available it is possible to fit to these points and orientations and thus estimate the surface. When the surface is known to be twice differentiable, a one parameter family of locally best fitting paraboloids is calculable from the stereo outlines. The parameter can then be estimated from shading information by solving a restricted version of the classical stereo correspondence problem.

In the case of generic algebraic surfaces, Bezout's Theorem predicts the number of frontier points in terms of the degree d of the surface. These can then be used to prove that (for surfaces with d>2) a stereo pair of outlines contain sufficient information to completely define the algebraic surface.

Back to index

Dr Anthony Di Pietro PhD 2007

Optimising Evolutionary Strategies for Problems with Non-uniform Noise

Supervisors: Dr Lyndon While and Dr Luigi Barone

Research Group: Adaptive Systems

Synopsis:

For many "real world" applications of evolutionary computation, the fitness function is obscured by random noise, which interferes with the evolutionary search. Furthermore, the amount of noise (noise strength) may vary throughout the search space, further complicating matters. Previous work has generally focussed on the specific case where noise strength is constant; however, we study problems with varying noise strength. We give new algorithms specifically designed to handle such problems, show how they perform, and provide a means to automatically apply them.

Back to index

Dr Adam Dunn PhD 2007

A New Model of Wildfire Prop

Postgraduate Student Profiles

Current Students and their Projects

Synopsis

The project focuses on different aspects of shape reconstruction from single and multiple image(s). It includes deriving novel algorithms (e.g. based on finite difference methods) for explicitly solving an unknown shape.

Synopsis

CSSE PhD and MSc Graduates