This thesis contains a study of state of the art knowledge acquisition modelling principles and methods for modelling general domain knowledge. This includes Newell's knowledge level, knowledge level modelling, Components of Expertise, CommonKADS and the Protégé meta tool. The thesis also includes a short introduction to the knowledge-intensive case-based reasoning system TrollCreek. Based on this background knowledge, one did analysis and comparison of different possible solutions. Then, after justifying the choices made, a knowledge acquisition method for TrollCreek was created. The method was illustrated through an example, evaluated and discussed.

Systems that efficiently provide support for adaptive work processes in a mobile environment do not exist according to our knowledge. Such systems would increase efficiency and safety in environments where work is inherently mobile, ad-hoc, and requires input from a set of heterogeneous sources. Traditional work support systems are normally not capable of dynamic change and plans must be made before work is started. This conflicts with most work processes, which are dynamic and where plans cannot be completely pre-defined. Current workflow systems are for the most part not capable of handling dynamic change in the workflow process execution. Those that do exist are geared more towards long-term adaptability of the workflow process and not towards in-situ planning of activities. In this report we provide an overview over current research related to adaptive workflow, activity theory, situated actions, and context-awareness. Then, we further explore the concept of adaptive workflow and context-awareness and how this can be implemented in a prototype workflow enactment service. A set of requirements for such a system is elicited from this exploration. We also provide a possible scenario for usage of adaptive context-aware workflow technology. From these requirements we have created an overall architecture that supports adaptive context-aware mobile work. Our focus within this architecture is on context-aware adaptive workflow systems. We finally present the design and implementation of a prototype application supporting context-aware adaptive mobile work processes. This prototype has been named PocketFlow and is implemented in embedded visual C++ for Microsoft PocketPC 2003 second edition.

Oppgaven har tatt for seg tre forskjellige måter å utvide en open hardware prosessor på. En av metodene har blitt brukt for en faktisk implementasjon, mens de to andre har blitt brukt til simulering. Hardware aksellereringen har blitt brukt for å kjøre MPEG-2 dekoding på prosessoren på en mer effektiv måte. Resultatene fra simuleringene har så blitt brukt for å se på klokkefrekvens, areal og strømforbruk til den nye SoCen/prosessoren.

Modern graphics processing units (GPUs) have evolved into high-performance processors with fully programmable vertex and fragment stages. As their functionality and performance are still increasing, more programmers are appealed by their computational power. This has led to an extensive usage of the GPU as a computational resource in general-purpose computing, and not just within applications of the entertainment business and computer games. Large volume data sets are involved when it comes to medical image segmentation. It is a time consuming task, but is important in the process of detection and identification of special structures and objects. In this thesis we investigate the possibility of using commodity graphics hardware for medical image segmentation. By using a high-level shading language, and utilizing state of the art technolgy like the framebuffer object (FBO) extension and a modern programmable GPU, we perform seeded region growing (SRG) on medical volume data. We also implement two pre-processing filters on the GPU; a median filter and a nonlinear anisotropic diffusion filter, along with a volume visualizer that renders volume data. In our work, we managed to port the Seeded Region Growing (SRG) algorithm from the CPU programming model onto the GPU programming model. The GPU implementation was successful, but we did not, however, get the desired reduction in time consume. In comparison with an equivalent CPU implementation, we found that the GPU version is outperformed. This is most likely due to the overhead associated with the setup of shaders and render-targets (FBO) while running the SRG. The algorithm has low computational costs, and if a more complex and sophisticated method is implemented on the GPU, the computational capacity and the parallelism of the of the GPU may be more utilized. Hence, a speed-up in computational time is then more likely to occur compared to a CPU implementation. Our work involving a 3D nonlinear anisotropic diffusion filter strongly suggests this.

Over the last few years, the computer graphics hardware has evolved extremely fast from supporting only a few fixed graphical algorithms to support execution of dynamic programs supplied by a developer. Only a few years back all graphics programs were written in assembly language, a nonintuitive low level programming language. Today such programs can be written in high level, near written English, source code, making it easier to develop more advanced effects and geometric shapes on the graphics card. This project presents a new way to utilize today's programmable graphics card to generate and render trees for real-time applications. The emphasis will be on generating and rendering the geometry utilizing the graphics hardware, trying to speed up the calculation of naturally advanced shapes for the purpose of offloading the systems central processing unit.

We have implemented and tested segmentation methods for segmenting brain tumours from magnetic resonance (MR) and ultrasound data. Our work in this thesis mainly focuses on active contours, both parametric (snakes) and geometric contours (level set). Active contours have the advantage over simpler segmentation methods that they are able to take both high- and low-level information into consideration. This means that the result they produce both depends on shape as well as intensity information from the input image. Our work is based on the results from an earlier completed depth study which investigated different segmentation methods. We have implemented and tested one simplified gradient vector flow snake model and four level set approaches: fast marching level set, geodesic level set, canny edge level set, and Laplacian level set. The methods are evaluated based on precision of the region boundary, sensitivity to noise, the effort needed to adjust parameters and the time to perform the segmentation. We have also compared the results with the result from a region growing method. We achieved promising results for active contour segmentation methods compared with other, simpler segmentation methods. The simplified snake model has given promising results, but has to be subject to more testing. Furthermore, tests with four variants of the level set method have given good results in most cases with MR data and in some cases with ultrasound data.

Implementation of automated volume-to-volume registration applications for three separate registration steps desired in enhancing neurosurgical navigation is considered. Prototype implementations for MRI-to-MRI registration, MRI-to-US registration and US-to-US registration have been made using registration methods available in the Insight Toolkit, with variants of the Mutual Information similarity metric. The obtained results indicate that automatic volume-to-volume registration using Normalized Mutual Information should be feasible for the neuronavigational applications considered here, with sufficient accuracy.

The task of this assignment is to make an algorithm that can detect movement regardless of how the illumination is. Handling changes in illumination is an important part of creating a stable surveillance system. This problem has been attempted solved here by making a model of the scene which consists of the expectation value and the standard deviation value for each pixel. For every frame that is tested for movement a ratio $p$ is calculated that is the relationship between the actual pixel value $x$, the expectation value $mu$ and the standard deviation value $sigma$. Three different methods were made that use these $p$ values to look for movement. The method that turned out to work best under all conditions compares the $p$ value for each pixel with the $p$ values of its neighbours. This solution is based on the observation that the relation between the greyscale values of pixels in a small area doesn't change. The system is tested both indoor and outdoor. It handles moving shadows and big changes in the illumination without triggering too many false alarms, and at the same time it detects movement under different illumination environments. When tested on a uniform scene it detected 87.7 % of the movement that was presented to the system. The hardest movement to detect were dark objects on a dark background. The system has a problem when the greyscale value of the moving object gets too similar to the greyscale value of the scene. And if the scene has some areas with monotonous texture and some areas with complex texture, the monotonous texture tends to get less sensitive. This problem is proposed solved by splitting the region of interest in several smaller areas, to make each area equally sensitive for movement.

Unmanned Aerial Vehicles (UAVs) have a tremendous appeal. One can imagine a large number of applications such as search-and-rescue, traffic monitoring, aerial mapping, etc. Helicopters are particularly attractive due to their Vertical Take Off and Landing (VTOL) capabilities. The research on UAVs has shown rapid development in recent years, and offers a great number of challenges. This thesis is the result of a project which is a part of the Autonomous Remote Controlled Helicopter (ARCH) project at the Department of Computer and Information Science, Norwegian University of Science and Technology. The ARCH project has already gained public interest, when it was featured on a television program (Schrödingers katt, NRK. September 2004). The object of this thesis is divided into three main sections. Firstly, it is to create and describe a remote control system for controlling the UAV in semi-autonomous mode, that will also enable the UAV to autonomously follow objects (pursuit-mode). Secondly, it is to create and describe a virtual cockpit which is to be used with the remote control system. Finally, it is to create and describe an image stabilization system, which can stabilize the visual information sent from the UAV to the ground and the virtual cockpit. These three components have been combined and integrated into the client prototype called ARCH Groundstation. Together, these three components provides a platform for an operator to control the ARCH UAV in semi-autonomous mode.

When doing image guided surgery, it is important to find a proper alig nment of the coordinate systems for the images and for the tracking system that tracks the positions o f the surgeons tools. This report explores surface based methods for finding such an alignment, using either an optical shape measurement device, or surfaces gathered by passing the tracking tool along the surface of the patien t. Accuracy and usability factors are explored, and compared to existing methods based on finding corresponding points