PhD

MSc

This thesis describes a system designed for facial expression control. A prototype in the Java language has been implemented based on this design. The target of facial animation is an imaginary character. Firstly, the geometric data of the head of a given character is read from a file. The file is in VRML format and the head geometry it contains is in a hierarchical structure. Then, the geometry of the head is distorted to get different facial expressions of emotion based on a pseudo-muscle system. These facial expressions are mapped to emotion. The result of this mapping is used to create a system for facial expression control by emotion. A character, Floops, has been chosen for illustrating the concepts being described. The pseudo-muscle system designed in this thesis enables Floops to exhibit a wide range of emotions on its face. The system allows us to control the facial expression of Floops by changing its emotion. The control is very easy to handle. When running on a computer with a moderate processor power (a Pentium 200MHz personal computer), the facial expression can be changed in real time, even when running the prototype on a Java virtual machine.

Currently, the majority of tools available for visual programming are designed to work in the two-dimensional paradigm where syntactic elements and data are represented by 2D graphical objects. This thesis presents the possibility of extending these tools to include 3D graphics as a medium for visual programming.

We describe the design and implementation of 3DComposer (3DC), a visual builder for composing visual notations with three-dimensional appearances. Individual notational elements, known as 3Dvixels, are created in an interactive 3D editing environment that supports the direct manipulation of objects. These 3Dvixels are defined using a combination of Java for functionality and the Virtual Reality Modelling Language (VRML) for visual appearances.

3Dvixels are exported as JavaBeans reusable software components that can be used in meta-tools, such as JComposer, for generating multi-view editing systems. These 3Dvixels are useful in developing applications for creating visual programming languages with 3D syntax and information visualisation tools that visualise data with 3D representations.

While volume rendering is a well researched topic, many existing renderers sacrifice accuracy for execution speed. The research in this thesis aims for quality over speed. We introduce a ray-casting method that retains fidelity to the emis-sion- absorption volume model, using a high accuracy integration approach. The integration formula handles piecewise linear emission and piecewise constant absorption, and correctly calculates the attenuation of light within each integration step, unlike many common methods.

For efficient rendering, a pointerless octree representation of the volume was used, with each ray performing a leaf by leaf traversal. Rays are able to traverse from leaf to leaf quickly without climbing the tree or performing a search, as in other methods.

We examine the quality of the renderer when used with a number of medical data sets, and discuss the advantages of a high quality rendering for image generation. We find that it is possible to generate high resolution images from reasonably sized data sets using the modest resources of current desktop personal computers. High order methods for integration prove to give greater accuracy for the same calcula-tion time as standard methods.

This thesis examines the feasibility of implementing a four-strip treadle configuration for the Indonesian Automatic Vehicle Classification (AVC) system. The Indonesian AVC system classifies vehicles as they pass through tollgates, without requiring any action by the driver or toll attendant. The configuration is intended to supplement the existing manual toll collection system.

There are three determinants in Indonesia's AVC system: axle distance, number of axles and height over first axle. In this thesis, we study how an AVC system might estimate axle distance by analysing the detection times of first and second axles on four-strip treadle. We conclude that axle distance can be estimated accurately under certain assumptions on measurement system resolution and motor vehicle technical specifications.

A population area of vehicle behaviour is set within acceptable ranges of axle distance, acceleration and velocity values. The accuracy of the axle distance estimator is examined by calculating the probability of misclassification over that population.

The results of this study show that all currently registered vehicles would be correctly classified, provided that the axle counter and the height sensor measuring the other two determinants are 100% accurate.

Soft objects are those everyday objects that deform significantly in response to their normal environment. Examples include cushions, plasticine and balloons filled with treacle. We present a new physically based model for animating soft objects. The model consists of two components: an elastic surface and a compressible fluid. The surface envelops the fluid, and is represented as a mass spring system. The fluid is modelled using finite difference approximations to the Navier-Stokes equations of fluid flow. Where the fluid and surface are in contact, their behaviour is defined by the no-slip condition. Unlike models that use implicit surfaces, surface tension is a natural feature of the model. Furthermore, the volume of the object is maintained depending on the properties of the fluid, and no explicit volume adjustment is necessary. We develop the model in one, two and three dimensions. We illustrate the behaviour of the soft object in environments with gravity and collisions with planes.