The purpose of this thesis was to examine ways to adapt common 2D segmentation techniques to work with 3D grids. The focus of the thesis became how to automate and improve the performance of region growing in 3D grids. After examining relevant literature and developing a tool to run experiments, a simple automatic region grower for 3D grids was developed. Quantitative performance measures and qualitative analysis of the segmentation results were performed. This algorithm was then used as a baseline for comparison when developing a more advanced region grower for 3D grids based on the seeded region grower (SRG) for 2D grids. This new algorithm was then modified to improve its speed and later extended to allow fully automatic operation by automating the placement of starting seeds. It was found that for the algorithms that were extended to a 3D grid, the main challenge was the resources needed by these algorithms when operating on high resolution grids. It was found that even though there have been steady and rapid improvements in consumer hardware since the original region growing algorithms were used on 2D grids, the very large amounts of data resulting from an extension from surface grids to volume grids requires that special attention is paid to handling resources effectively. It was further revealed that what was considered the best data structures and algorithms for the SRG algorithm when it was first introduced, is not necessarily the best choice on todays computing hardware. Also, the conclusion is drawn that with regards to performance, it is now possible to segment volumes approximately as fast as surfaces were segmented in the early 1990s.

Evolutionary artificial neural networks can adapt to new circumstances, and handle slight changes without catastrophic failure. However, under constantly changing circumstances, resulting in unpredictable grounds for evaluating success, the lack of memory of previous adaptations are a limiting factor. While further evolution can allow adaptations to new changes, the same is required for a return to a previous environment. To reduce the need for further evolution to deal with previously seen problems, this thesis looks at an approach to encourage previous knowledge to be retained across generations. It does this using back propagation in conjunction with an implementation of the HyperNEAT neuroevolutionary algorithm.

Many problems in the field of automatic video surveillance exists today. Some have yet to be overcome. One of these problems is how a computer system automatically can determine if a situation should cause an alarm or not. To resolve this problem, the use of Case-based reasoning (CBR) is proposed. CBR is a technique that allows a system to reason about different situations and to learn from them. The aim is to produce a system that utilizes these abilities. The system should learn to recognize the situations that causes different alarms. When a situation is recognized and categorized, these false alarms can be completely avoided. This master thesis explains and shows the advantages of using such a system together with advanced image processing techniques.

Task-based programming involves creating tasks, which can be run independently of each other, and letting the run-time system schedule the tasks on the underlying architecture. Wool is a new library for task-based programming created at SICS in Sweden. To assist a developer who is using Wool to parallelize a program, as well as the scientists who are actually developing Wool, a profiler which shows what happened in a computation can be very helpful.In this project we modify the Wool library to print more data about its computations. When the output is given to a Java application also developed in this project, the Java application produces a graphical representation of the execution. Each worker thread is visualized separately, with spawns, steals, leaps, critical path and CPU usage information included at a position corresponding to when the events actually occurred.The profiler, which we have named WoolPlot, is put to the test using a few real-world benchmarks, as well as some created especially for this project. The benchmarks show that WoolPlot works well when describing the distinct events such as steals and spawns. The reporting on the CPU load is too inaccurate to be sufficient for all practical uses. The overhead of the profiler is estimated to be between 3% and 6%.

Using the parallel architecture of the graphics processing unit for general purpose programming has become increasingly common in the recent years. The process of creating a mathematically correct transformation of a scene for curved stereoscopic projection is a very expensive task, which would greatly benefit from a massively parallel solution implemented on the GPU.In this thesis, we first investigate two different methods for obtaining a mathematically correct transformation of images intended for stereoscopic projection on strongly curved surfaces. One method revolves around transforming a pre-rendered image, pixel by pixel, while the other method applies the transformation to the projection of the vertices in the scene before they are rendered as an image. We then develop massively parallel solutions for both these methods on the GPU, striving to a reach a real-time rate for the stereoscopic projection of the transformed images.We test both methods for different problem areas, and compare the results to map their strengths and weaknesses. From the obtained results, we conclude that they are both useful in different areas. The vertex transformation performs poorly when the number of vertices in the scene is very high, but for a moderate number of vertices it achieves excellent results, even for exceptionally large image resolutions. The pixel transformation is far less affected by the number of vertices in the scene; however its performance declines rapidly as we increase the size of the image. Both methods were able to execute in real-time for relevant problem sizes.

In this report I propose a method of navigation for differentially wheeled robots inspired by path integration in certain social insects like bees and ants. It is a very simple method, intended for use in low-tech robots with very limited hardware, such as swarm robots. Path integration is essentially dead reckoning as used by animals, calculating the relative position based on the movements made since the last known position. It is a tried and true method of navigation that also has significant flaws, especially in that inaccuracies accumulate and magnify over time. In this report I want to examine whether communication and information sharing between robots in a swarm can alleviate some of the drawbacks, and make it a viable method for navigation for swarm robots over relatively short distances.

In the later years robotics has seen a huge increase within domestic use, and have now become an affordable tool in the daily life of most people. The goal of this project was to investigate the differences between a physical and virtual robot in terms of increased content knowledge, learning motivation, and interest in science, technology, engineering and mathematics (STEM). To investigate this we conducted an experiment at Trondheim’m International School (THIS), using a quasi-experimental setup with two treatment group, virtual and physical robot. The results showed that there does not exist a statistically significant difference in content knowledge gain, motivation or interest between the robotics group and the simulator group.

In the later years robotics has seen a huge increase within domestic use, and have now become an affordable tool in the daily life of most people. The goal of this project was to investigate the differences between a physical and virtual robot in terms of increased content knowledge, learning motivation, and interest in science, technology, engineering and mathematics (STEM). To investigate this we conducted an experiment at Trondheim’m International School (THIS), using a quasi-experimental setup with two treatment group, virtual and physical robot. The results showed that there does not exist a statistically significant difference in content knowledge gain, motivation or interest between the robotics group and the simulator group.

The goal of the project in this thesis is to explore the viability of an approach to user adaptation where the application context is significantly more constraining than in most cases seen in previous academic work.The project describes and implements a system for rolling out product features incrementally in an optimal way, based on feature adoption statistics within user segments. In other words, the described system should allow for simple personalization of the product.When analyzing the adoption rate of new features, we find that there are indeed clear differences between the identified user segments. However, limitations due to the lack of stable user identity make adaptation approaches based on these data unfeasable in practice.

This paper presents an algorithm to solve a variant of the bin packing problem with additional constraints on the order of items. The performance of this algorithm is tested, both for optimal solutions and approximations given by early termination, and is found to be limited for optimal solutions, but fairly efficient for decent approximations.