Thèses sur le sujet « Phase-Based Algorithms »

Cette thèse présente deux contributions dans le domaine des systèmes distribués : OMAHA et le Consensus f-Révocable. OMAHA (Opportunistic Message Aggregation for pHase-based Algorithms) est un mécanisme d'agrégation de messages conçu pour les algorithmes à phases. Dans les environnements Cloud, où plusieurs applications partagent la même infrastructure, la bande passante est une ressource critique. Dans les datacenters, une partie importante du trafic est consacrée aux petits messages. Chaque message possède un en-tête, ainsi, l'accumulation de ces petits messages pose un véritable problème et entraîne une consommation non négligeable de bande passante. Plusieurs mécanismes ont été proposés pour adresser ce défi, mais peu d'entre eux prennent en compte les caractéristiques applicatives. Ils reposent principalement sur une agrégation des messages au niveau de la couche réseau. OMAHA exploite les spécificités des algorithmes à phases pour agréger intelligemment et de manière opportuniste les messages. Bien que gourmands en messages, de très nombreuses applications reposent sur des algorithmes à phases (Google Spanner, Zookeeper, etc.). Cependant, ils possèdent un avantage majeur : des communications prévisibles. En anticipant les futures communications, OMAHA retarde les messages en les regroupant avec d'autres destinés au même processus. Cela permet de réduire le nombre de messages envoyés sur le réseau et donc une économie de bande passante. Nos expériences montrent des économies de bande passante allant jusqu'à 30%, tout en limitant la dégradation de la latence à 5% pour le célèbre algorithme de Paxos. Dans les systèmes distribués, atteindre un consensus sur une action ou une valeur est un défi complexe, surtout lorsque les processus sont soumis à des contraintes. De nombreux systèmes, tels que les systèmes multi-agents (véhicules autonomes, gestion d'agenda, robotique, etc.) ou encore les systèmes d'allocation de ressources, doivent respecter des contraintes tout en atteignant un objectif global commun. Cependant, les algorithmes de consensus traditionnels ne prennent pas en compte ces contraintes, la décision se basant uniquement sur les valeurs proposées par les processus. Une solution triviale consisterait à recueillir toutes les contraintes, mais en raison de l'asynchronisme et des pannes, cela est impossible. Pour tolérer les pannes, certains algorithmes définissent un nombre maximal de fautes qu'ils peuvent supporter. Cela permet à l'algorithme de progresser sans attendre la réponse de tous les processus. En conséquence, la valeur décidée est souvent imposée par un sous-ensemble de processus, la majorité. Les contraintes de la minorité sont ainsi ignorées. Pour répondre à ce problème, nous avons introduit le Consensus f-Révocable. Ce consensus revisité permet de choisir une valeur qui respecte les contraintes des processus tout en offrant la possibilité de révoquer une décision prise par la majorité si celle-ci viole les contraintes d'un processus de la minorité. La convergence est assurée car le nombre de révocations est borné par le nombre de processus appartenant à la minorité. Nous avons développé deux adaptations de l'algorithme Paxos pour mettre en pratique ce nouveau consensus
This thesis presents two contributions to the field of distributed systems: OMAHA and the f-Revoke Consensus.OMAHA (Opportunistic Message Aggregation for pHase-based Algorithms) is a message aggregation mechanism designed for phase-based algorithms. In cloud environments, multiple applications share the same infrastructure, making bandwidth a critical resource. A significant portion of traffic in data centers consists of small messages. Each message includes a header, leading to substantial bandwidth consumption. Several mechanisms have been proposed to address this issue, but few consider application-specific characteristics. Most rely on aggregation at the network layer. OMAHA leverages the features of phase-based algorithms to intelligently and opportunistically aggregate messages. Many applications, such as Google Spanner and Zookeeper, depend on phase-based algorithms. They are often message-intensive but offer a key advantage: predictable communications. By anticipating future communications, OMAHA delays messages and groups them with others intended for the same process. This approach reduces the number of messages sent over the network, resulting in bandwidth savings. Our experiments show bandwidth saving of up to 30%, while limiting latency degradation to 5% for the well-known Paxos algorithm. In distributed systems, achieving consensus on an action or value is complex, especially when processes face constraints. Many systems, including multi-agent systems (like autonomous vehicles and robotics) and resource allocation systems, need to respect these constraints while working towards a common goal. Unfortunately, traditional consensus algorithms often overlook these constraints, focusing only on the values proposed by the processes. A straightforward solution would be to gather all constraints, but due to asynchrony and potential failures, this is impossible. To handle failures, some algorithms set a limit on the number of faults they can tolerate. This allows them to move forward without waiting for responses from every process. As a result, the final decision is made by a subset of processes known as the majority. This leads to the exclusion of constraints from the minority. To tackle this problem, we introduced the f-Revoke Consensus. This new approach enables the selection of a value that considers processes' constraints. It also allows for the revocation of a majority decision if it violates the constraints of a minority process. Importantly, convergence is ensured because the number of revocations is limited by the size of the minority. We developed two adaptations of the Paxos algorithm to implement this new consensus

Dissertação (mestrado) - Universidade Federal de Santa Catarina, Centro Tecnológico, Programa de Pós-Graduação em Engenharia Mecânica, Florianópolis, 2014.
Made available in DSpace on 2015-02-05T21:16:46Z (GMT). No. of bitstreams: 1 331656.pdf: 3739147 bytes, checksum: 4434e631824c3243a2eb1e0e7148fe81 (MD5) Previous issue date: 2014
O correto funcionamento de tubulações subterrâneas para o transporte de gás e petróleo depende de um monitoramento frequente e correto dos estados de tensões. Avanços recentes na medição de tensões residuais têm aplicado métodos ópticos em conjunto com o alívio de tensões de maneira a avaliar o campo de tensões no componente. Estes métodos requerem uma etapa de remoção do salto de fase para interpretar corretamente os dados adquiridos. Esta remoção do salto de fase tem sido um desafio para diversas aplicações metrológicas. Este trabalho tem por objetivo propor uma abordagem original para a solução deste problema. Neste trabalho é apresentado o algoritmo proposto assim como diversos resultados com diferentes imagens comparados com métodos consagrados.A luz, comportando-se como onda, obedece ao princípio de superposição que por sua vez dá lugar ao fenômeno de interferência. Este fenômeno pode ser utilizado de diversas maneiras para a medição de superfícies e formas geométricas. No entanto, várias dessas aplicações, como interferometria speckle e shearografia, fornecem os valores de interesse restringidos a um intervalo de ?p a p. Assim, faz-se necessária uma operação para retomar os valores reais que produziram o resultado obtido. Esta operação é chamada de remoção do salto de fase.Por décadas tem-se estudado diversas técnicas para realizar a remoção do salto de fase. Elas podem ser divididas em duas categorias principais: métodos que seguem caminhos e métodos independente de caminhos. Métodos que seguem caminhos aplicam uma simples equação de comparação e adição de múltiplos de 2p por toda a imagem. Elas diferem nos caminhos de pixels escolhidos. Para que o resultado seja confiável, é necessário que esse caminho evite pixels de baixa qualidade ou corrompidos. As técnicas de branch-cut identificam esses pixels através da teoria de resíduos e conectando resíduos de sinais opostos, ela é capaz de traçar caminhos confiáveis para a remoção do salto de fase. Técnicas baseadas em qualidade atribuem notas relativas a diferentes critérios de qualidade para cada pixel, excluindo da análise aqueles que se encontram abaixo de um limiar arbitrário.Técnicas independentes de caminhos, como os métodos de norma mínima, assemelham-se a métodos de otimização. Estes são iterativos e procuram por um mínimo na diferença entre as derivadas da solução proposta e as derivadas da imagem original. Estes métodos são considerados bastante robustos e confiáveis. No entanto, estes tambémdemandam maior tempo de processamento para encontrar a resposta correta.Em paralelo aos desenvolvimentos na área de remoção do salto de fase, cientistas têm desenvolvido técnicas computacionais baseadas no comportamento de animais sociais. O campo de Inteligência de Enxame é inspirado por insetos como formigas, abelhas e cupins e outros animais como peixes e pássaros. Estes animais têm em comum o fato de criarem sistemas organizados embora compostos de elementos simples e a ausência de uma liderança clara. O comportamento de formigas e abelhas na busca por comida e os movimentos em grupo de peixes e pássaros são os exemplos mais claros do conceito de comportamento emergente: um comportamento que, embora não explícito na descrição de seus elementos individuais, surge com a interação entre diversos desses elementos. Este comportamento emergente pode ser explicado em termos de agentes simples e independentes, regras simples e um comportamento descentralizado.Este fenômeno tem inspirado as ciências da computação por décadas. Diversas soluções computacionais para problemas matemáticos ou operacionais têm sido propostas a partir das soluções elegantes encontradas na natureza. Exemplos dessas soluções são os algoritmos de otimização baseados no comportamento de formigas e abelhas. No entanto, pouco deste conceito tem sido aplicado na área de processamento de imagem. Quanto ao problema de remoção do salto de fase, mais especificamente, não foi encontrado nenhum trabalho que propusesse uma solução baseada em Inteligência de Enxame.Assim, o presente trabalho propõe uma solução baseada nestes conceitos. Por causa da natureza imprevisível do comportamento emergente, o desenvolvimento do algoritmo proposto foi pouco convencional. Em primeiro lugar, foi necessário o desenvolvimento de um ambiente de testes onde o enxame pudesse ser observado em tempo real durante a sua operação. Em segundo lugar, a criação do algoritmo se deu de maneira iterativa até que fosse encontrado um conjunto de regras satisfatório.Uma primeira solução foi encontrada modelando os agentes como máquinas de estados finitos. Este modelo de agente foi implementado com dinâmicas de comunicação indireta através de estigmergia e comunicação direta em casos de necessidade. Este método, apesar de ter apresentado bons resultados em termos de qualidade da remoção do salto de fase, necessitava ainda de um critério de parada independente do usuário. Na criação deste critério de parada, novas regras deram espaço para a criação de um algoritmo completamente diferente.Esta segunda solução modela o agente a partir de cinco regras simples que permitem, entre outras coisas, a criação e desativação de novos agentes. Uma vez que todos os agentes são desativados, o programa chega ao fim e retorna a imagem com o salto de fase removido. A primeira destas regras afirma que se há um ou mais pixels que podem ter seu salto removido na vizinhança do agente, um deles será escolhido aleatoriamente para a operação. O agente então se move para o pixel escolhido e ganha um ponto de energia. Se não há pixels aptos a serem trabalhados, um pixel já trabalhado na vizinhança é escolhido aleatoriamente, de acordo com a segunda regra. O agente se move para o pixel escolhido e perde um ponto de energia. A terceira regra faz com que agentes que encontram dois pixels vizinhos já trabalhados mas inconsistentes entre si, marcarem estes pixels como defeituosos e desativarem-se. As duas últimas regras fazem com que agentes com energia excedente repliquem-se e aqueles sem energia desativem-se.O comportamento esperado é que os agentes de distribuam pela imagem de maneira eficiente, aproveitando ao máximo os ciclos de processamento. Além disso, a regra de marcação de remoções duvidosas faz com que problemas de ambiguidade na remoção do salto de fase não sejam propagados por grandes regiões da imagem. Este algoritmo foi testado em diversas condições e comparado com outros métodos estabelecidos.Os primeiros resultados foram gerados aplicando-se o enxame em imagens sintéticas sem quaisquer erros. Assim, foi possível avaliar a influência de diferentes parâmetros escolhidos pelo usuário no comportamento do enxame e qualidade dos resultados. Foi possível observar o impacto dos parâmetros de energia na densidade do enxame que, por sua vez, é importante para a correção de ambiguidades propagadas.Em seguida, foram testadas imagens sintéticas com erros artificiais. Os resultados foram comparados com um algoritmo baseado em qualidade e um algoritmo de norma mínima. Foi observado que o algoritmo proposto foi extremamente capaz de contornar as dificuldades das imagens de maneira, produzindo resultados confiáveis. Para certas condições, os resultados foram ainda melhores que os obtidos pelo outro algoritmo baseado em qualidade.Foram testadas ainda imagens provenientes de aplicações metrológicas reais: projeção de franjas, interferometria speckle e shearografia. Os resultados obtidos pelo algoritmo baseado em Inteligência de Enxame foram bastante satisfatórios, comparáveis aos métodos mais robustos. Ainda, o algoritmo proposto apresentou melhoresresultados para imagens muito ruidosas quando comparado com o outro algoritmo baseado em qualidade testado. Estes resultados atestam do potencial do método proposto em obter resultados rápidos e confiáveis.Por fim, este trabalho foi concluído com um breve resumo destes resultados e a validação dos objetivos originais, afirmando assim o sucesso do método proposto. Foram listadas ainda algumas sugestões para avanços futuros como os testes com imagens e parâmetros de qualidade novos, a implementação de processamento paralelo e a criação de novas abordagens baseadas em Inteligência de Enxame para a solução deste problema e outros semelhantes.

Abstract : The proper functioning of underground oil and gas pipelines depend on the frequent and correct monitoring of stress states. Recent developments on residual stress measurement techniques have employed optical methods allied with stress relief in order to assess the underlying stress field. These optical methods require a phase unwrapping step to interpret the acquired data correctly. Phase unwrapping has posed a challenge for many optical metrology applications for decades and saw the development of many different solutions. For the past decades, the field of Swarm Intelligence, based on the behavior observed among ants, bees and other social insects, has been studied and many algorithms have been designed to perform a variety of computational tasks. Swarm Intelligence is commonly regarded as robust and fast, which are desirable features in a phase unwrapping algorithm. This work proposes a novel approach to phase unwrapping based on Swarm Intelligence, assessing its applicability, comparing it to existing methods and evaluating its potential to future developments. The proposed algorithm is thoroughly explained and the results for several different images are presented. These results show a great potential of the proposed method, performing better than some established techniques in specific situations. This potential is assessed and suggestion for future advancements are given.

Distributed Generators (DGs) are sensible to voltage sags, so the protection devices must trip fast to disconnect the faulted part of the grid. The DG disconnection will not be desirable in the near future with a large penetration, so it will be necessary to lay down new requirements that should be based on avoiding unnecessary disconnections. Therefore, to prevent unnecessary tripping when inverter-based DGs are connected to the Medium Voltage (MV) grid, reliable and effective protection strategies need to be developed, considering the limited short-circuit current contribution of DG. The initial goal of this study is to employ different possible control strategies for a grid-connected inverter according to the Spanish grid code and to analyze the output voltage behavior during symmetrical and unsymmetrical voltage sags. The analytical development of the proposed strategies shows the impacts of the sag on currents, voltages, active and reactive powers. Another goal of this research is to propose a protection strategy based on Artificial Intelligence for a radial or ring distribution system with high DG penetration. The protection strategy is based on three different algorithms to develop a more secure, redundant, and reliable protection system to ensure supply continuity during disturbances in ring and radial grids without compromising system stability. In order to classify, locate and distinguish between permanent or transient faults, new protection algorithms based on artificial intelligence are proposed in this research, allowing network availability improvement disconnecting only the faulted part of the system. This research introduces the innovative use of directional relay based on a communication system and Artificial Neural Network (ANN). The first algorithm, Centralize algorithm (CE), collects the data from all the PDs in the grid in the centralized controller. This algorithm detects the power flow direction and calculates the positive-sequence current of all the PDs in the grid. Significant benefits of this system are that it consolidates the entire systems security into a single device, which can facilitate system security control. However, the CE will not pinpoint the exact location of the fault if there is any loss of information due to poor communication. Therefore, the systems redundancy can be improved by cooperating with a second algorithm, the Zone algorithm (ZO). ZO algorithm is based on zone control using peer-to-peer connectivity in the same line. The faulty line in that zone may be identified by combining the two PDs data on the same line. The most relevant advantage of this algorithm is its flexibility to adapt to any grid modification or disturbance, even if they are just temporary, unlike the CE, which is fixed to the existing grid configuration. The third protection algorithm, Local algorithm (LO), has been proposed without depending on the communication between the PDs; then, the protection system can work properly in case of a total loss of communication. Each PD should be able to detect if the fault is located in the protected line or another line by using only the local information of the PD. According to the type of fault and based on local measurements at each PD of abc voltages and currents, different algorithms will be applied depending on the calculation of the sequence components. The main advantage of this algorithm is the separate decision of each PD, and avoiding communication problems. In case of radial grids, both mechanical breakers and Solid State Relays (SSRs) are used to verify the protection strategies, and in the case of ring grids, mechanical breakers are used, due to the limitations in required voltage difference of SSR. The proposed protection algorithms are compared with conventional protections (Overcurrent and Differential) protections to validate the contribution of the proposed algorithms, especially in reconfigurable smart grids.
El objetivo inicial de este estudio es emplear diferentes estrategias de control posibles para un inversor conectado a la red segun el código de red español y analizar el comportamiento de la tensión de salida durante caídas de tensión simétricas y asimétricas. El desarrollo analítico de las estrategias propuestas muestra los impactos de los huecos de tensión en las corrientes, tensiones, potencias activas y reactivas. Otro objetivo de esta investigación es proponer una estrategia de protecclón basada en lnteligencia Artificial para una red del Sistema de Distribución, radial o en anillo, con elevada penetración de Generación Distribuida. La estrategia de protección se basa en tres algoritmos diferentes para desarrollar un sistema de protección más seguro, redundante, y fiable, que asegure la continuidad de suministro durante perturbaciones en redes radiales o en anillo sin comprometer la estabilidad del sistema. Para clasificar, localizar y distinguir entre faltas permanentes o transitorias, se proponen en este trabajo nuevos algoritmos de protección basados en inteligencia artificial, permitiendo la mejora de la disponibilidad de la red, al desconectar sólo la parte del sistema en falta. Esta investigación introduce la innovación del uso del rele direccional basado en un sistema de comunicación y Redes Neuronales Artificiales (ANN). El primer algoritmo, Algoritmo Central (CE), recibe los datos de todos los PDs de la red en un control central. Este algoritmo detecta la dirección de flujo de cargas y calcula la corriente de secuencia positiva de todos los PDs de la red. El entrenamiento de ANNs incluye variaciones en la corriente de cortocircuito y la dirección del flujo de potencia en cada PD. Los beneficios mas significativos de este sistema son que concentra la seguridad total del sistema en un único dispositivo, lo que puede facilitar el control de la seguridad del sistema. Sin embargo, el CE no determinara con precisión la localización exacta de la falta si hay alguna perdida de información debida a una pobre comunicación. Por lo tanto, la redundancia del sistema se puede mejorar cooperando con un segundo algoritmo, el algoritmo de Zona (ZO). El algoritmo ZO se basa en un control de zona usando la conectividad entre dispositivos de protección de una misma línea. La línea en falta en esa zona puede identificarse combinando los datos de los dos PDs de la misma línea.. La ventaja mas relevante de este algoritmo es su flexibilidad para adaptarse a cualquier modificación de la red o perturbación, incluso si sólo son temporales, a diferencia del CE, que se ha adaptado para la configuración de la red existente. El tercer algoritmo de protección, algoritmo Local (LO), ha sido propuesto sin dependencia de la comunicación entre PDs; por lo tanto, el sistema de protección puede operar correctamente en el caso de una pérdida total de comunicación. Cada PD debe poder detectar si la falta esta ubicada en la línea protegida o en otra línea, utilizando sóIo la información local del PD. Según el tipo de falta, y en base a medidas locales en cada PD, de tensiones y corrientes abc, se aplican diferentes algoritmos en función del cálculo de las componentes simétricas. La principal ventaja de este algoritmo es la actuación por separado de cada PD, evitando los problemas de comunicación. En el caso de las redes radiales, se utilizan tanto interruptores mecánicos como réles de estado sóIido (SSR) para verificar las estrategias de protección, y en el caso de las redes en anillo se utilizan interruptores mecánicos, debido a las limitaciones de tensión para su conexión. Los algoritmos de protección propuestos se comparan con protecciones convencionales (Sobrecorriente y Diferencial) para validar la contribución de los algoritmos propuestos, especialmente en redes inteligentes reconfigurables.
Enginyeria Elèctrica

Word processed copy.
Includes bibliographical references (leaves 71-72)
In this thesis, a new phase-based algorithm is developed, which overcomes the shortcomings of the Bollen algorithms. The new algorithm computes the dip type based on the difference in phase angle between the measured voltages.

The Internet Based real-time GPS synchronized wide-area Frequency Monitoring Network (FNET) is an extremely low cost and quickly deployable wide-area frequency measurement system with high dynamic accuracy which consists of Frequency Disturbance Recorder (FDR) distributed to more than 100 places around North America and an Information Management System situated at Virginia Tech. Since its first FDR deployment in 2003, the FNET system has been proved to be able to reliably receive phasor data accurately measured at and instantaneously sent via the Internet from different locations of interest, and efficiently run the analyzing program to detect and record significant system disturbances and subsequently estimate the location of disturbance center, namely the event location, in the electric grid based on the information gathered. The excellent performance of the FNET system so far has made power grid situation awareness and monitoring based on distribution level frequency measurements a reality, and thus advances our understanding of power system dynamics to a higher level and in a broader dimensionality. Chapter 1 and Chapter 2 of this dissertation briefly introduce the genesis and the architecture of the FNET system, followed by a summary of its concrete implementations. Chapter 3 and Chapter 4 outline FNET frequency estimation algorithm and phase angle estimation algorithm, including their attributes and the new methodologies to enhance them. In Chapter 5, the report discusses the algorithms developed at FNET to detect the frequency disturbance and estimate the disturbance location by the triangulation procedure using real-time frequency data and geographic topology of the FNET units in the power grid where the disturbance occurs. Then, the dissertation proceeds to introduce the FNET angle-based power system oscillation detection and present some research about Matrix Pencil Modal Analysis of FNET phase angle oscillation data in the following two chapters. Lastly, the content of this report is summarized and the future work envisioned in Chapter 8.
Ph. D.

Service Oriented Architecture (SOA) allows the development of software with requirements of interoperability and loose coupling. New services can be offered through the composition of existing services in order to address new business requirements. Business-to-business collaborative web applications demand a loosely coupled, coarse-grained interaction model. Thus, the traditional transaction architectures based on the ACID must be re-visited to support long-running web services. For such web services, it is not advisable to lock resources because they become blocked to other transactions. The two most adopted SOA architecture styles are the WS-* and REST (Representational State Transfer). WS-* is a set of specifications for the development of services based on SOAP and WSDL. The WS-* specifications support the implementation of non-functional requirements among them the transaction control. Two of the WS-* drawbacks are the implementation complexity and the strong dependency on SOAP and WSDL. REST, however, as an architectural style, does not provide "official" standards to address the non-functional requirements of services. We investigate the means by which a web service application can support transactions implementation using timestamp. We propose a protocol based in timestamp and two phase commit protocol in order to support the transactional control implementation in the web service domain. A protocol extension to improve the satisfaction of business rules is also proposed. The extension takes into account the application domain business rules which guide the web services behavior. In order to show the feasibility of the protocols, examples of an application are described. We also describe how the proposed protocols must be used to deal with host and connection failures.

State of the art direction finding (DF) systems usually have several antennas in order to increase accuracy and robustness to certain factors. In this thesis, a three antenna DF system is built and evaluated. While more antennas give better DF performance, a three antenna system is useful for system simplicity and many of the problems in DF systems can be observed and evaluated easily. This system can be used for both azimuth and elevation direction of arrival (DOA) estimation. The system is composed of three monopole antennas, an RF front end, A/D converters and digital signal processing (DSP) units. A number of algorithms are considered, such as, three channel interferometer, correlative interferometer, LSE (least square error) based correlative interferometer and MUSIC (multiple signal classification) algorithms. Different problems in DF systems are investigated. These are gain/phase mismatch of the receiver channels, mutual coupling between antennas, multipath signals and multiple sources. The advantages and disadvantages of different algorithms are outlined.

Phase-contrast imaging (PCI) is a modality of medical x-ray imaging that can solve one of the main limitations with conventional attenuation-based imaging: the imaging of materials with low attenuation coefficients, such as soft tissues. A modality of PCI, Propagation-based phase-contrast imaging (PBI), was used in this project. This method does not require any optical elements than those used in the conventional imaging; it does, however, require more processing compared to other kinds of PCI. In addition to the reduced image quality, the required image reconstruction process, with PCI, also requires several manual adjustments, which in turn results in a lot of time consuming. In order to achieve that, a simple iterative image reconstruction method that combines Simultaneous Iterative Reconstruction Technique (SIRT) and propagation-based phase-contrast imaging was developed. The proposed method was compared with another commonly used phase-retrieval method, Paganin's algorithm. The obtained results showed higher resolution and reduced blur artefacts compared with Paganin's method. The developed method also appeared to be less sensitive to error in the input parameters, such as the attenuation coefficient, but also more time-consumption than the non-iterative Paganin's method, due to the higher data processing.
Faskontrastavbildning är en ny medicinsk röntgenavbildningsteknik, som har utvecklats för att ge bättre kontrast än konventionell röntgenavbildning, särskilt för objekt med låg attenuationskoefficient, såsom mjuk vävnad. I detta projekt användes s.k. propagationsbaserad faskonstrantavbildning, som är en av de enkla metoder som möjliggör faskontrastavbildningen, utan extra optiska element än det som ingår i en konventionell avbildning. Metoden kräver dock mer avancerad bildbehandling. Två av de huvudsakliga problemen som oftast uppstår vid faskontrastavbildning är minskad bildkvalité efter den väsentliga bildrekonstruktionen, samt att den är tidskrävande p.g.a. manuella justeringar som måste göras. I det här projektet implementerades en enkel metod baserad på en kombination av den iterativa algoritmen för bildrekonstruktion, Simultaneous Iterative Reconstruction Technique (SIRT), med propagationsbaserad faskonstrantavbildning. Resultaten jämfördes med en annan fasåterhämtningsmetod, som är välkänd och ofta används inom detta område, Paganinsmetod. Efter jämförelsen konstaterades att upplösningen blev högre och artefakter som suddighet reducerades. Det noterades också att den utvecklade metoden var mindre känslig för manuell inmatning av parametern för attenuationskoefficient. Metoden visade sig dock vara mer tidskrävande än Paganin-metoden.

This Master Thesis discusses the Phase-II upgrade of the Trigger and Data Acquisition system of the detector called A Toroidal LHC ApparatuS (ATLAS). An R&D program, which has included the creation of two task forces, was launched in the 2021 Spring with the aim to produce one engineered solution for the track reconstruction at the Event Filter (EF) level. The Electronic group of the University of Bologna is taking part in the proposal of the heterogeneous commodity task force, which consists of the previous project called Hardware Tracking for the Trigger. The heterogeneous solution is based on a mixed commodity platform of classic processors and accelerators, where track reconstruction is expected to be performed via the use of mathematical functions, for example through the implementation of Hough Transform algorithm on the FPGA which will be part of the EF of ATLAS Phase-II. The R&D activities performed until now include the development of the firmware for the HT. Final goal of this Master Thesis is the creation of a hardware demonstrator able to test the ongoing firmware design. To fulfill this purpose, a new firmware architecture is exploited and it relies on a manageable Peripheral Component Interconnect Express transmission. The integration of the two firmware designs is realized with the development of a two first-in-first-out structure. In this way it is demonstrated the correct implementation of the ongoing Hough Transform firmware design.

Doutoramento em Engenharia Electrotécnica
Over the years, the increased search and exchange of information lead to an increase of traffic intensity in todays optical communication networks. Coherent communications, using the amplitude and phase of the signal, reappears as one of the transmission techniques to increase the spectral efficiency and throughput of optical channels. In this context, this work present a study on format conversion of modulated signals using MZI-SOAs, based exclusively on all- optical techniques through wavelength conversion. This approach, when applied in interconnection nodes between optical networks with different bit rates and modulation formats, allow a better efficiency and scalability of the network. We start with an experimental characterization of the static and dynamic properties of the MZI-SOA. Then, we propose a semi-analytical model to describe the evolution of phase and amplitude at the output of the MZI-SOA. The model’s coefficients are obtained using a multi-objective genetic algorithm. We validate the model experimentally, by exploring the dependency of the optical signal with the operational parameters of the MZI-SOA. We also propose an all-optical technique for the conversion of amplitude modulation signals to a continuous phase modulation format. Finally, we study the potential of MZI-SOAs for the conversion of amplitude signals to QPSK and QAM signals. We show the dependency of the conversion process with the operational parameters deviation from the optimal values. The technique is experimentally validated for QPSK modulation.
Nos últimos anos, a crescente procura e troca de informação tem levado ao aumento de tráfego nas redes de comunicação óticas atuais. As comunicações coerentes, com recurso à amplitude e fase do sinal, ressurgem como uma das técnicas de transmissão capazes de aumentar a eficiência espectral e o rendimento dos canais óticos. Nesse âmbito, este trabalho apresenta um estudo sobre a conversão de formatos de modulação de sinais, usando técnicas exclusivamente no domínio ótico, através de conversão de comprimento de onda, com base no MZI-SOA. Esta técnica, aplicada em nós óticos que interligam redes óticas com débitos binários distintos, permite uma maior escalabilidade e eficiência da rede. A tese começa por apresentar uma caracterização experimental detalhada das propriedades estáticas e dinâmicas do MZI-SOA. É depois proposto um modelo semi-analítico que descreve a evolução da amplitude e fase do sinal ótico à saída do MZI-SOA. Os coeficientes do modelo são obtidos recorrendo a um algoritmo genético multiobjectivo. O modelo é validado experimentalmente, explorando a dependência do sinal ótico com os parâmetros operacionais do MZI- SOA. Segue-se a proposta de uma técnica de conversão de formato de modulação de amplitude para modulação de fase contínua. Finalmente, é feito um estudo das potencialidades do MZI-SOA para conversão de formato de modulação de amplitude para modulação QPSK e QAM. Mostra-se a dependência da constelação do sinal com o desvio dos parâmetros operacionais, em torno do valor ótimo. A técnica é validada experimentalmente para modulação QPSK.

Image segmentation is one of the most difficult tasks in image processing. Segmentation algorithms are generally based on searching a region where pixels share similar gray level intensity and satisfy a set of defined criteria. However, the segmented region cannot be used directly for partial image retrieval. In this dissertation, a Contour Based Image Structure (CBIS) model is introduced. In this model, images are divided into several objects defined by their bounding contours. The bounding contour structure allows individual object extraction, and partial object matching and retrieval from a standard CBIS image structure. The CBIS model allows the representation of 3D objects by their bounding contours which is suitable for parallel implementation particularly when extracting contour features and matching them for 3D images require heavy computations. This computational burden becomes worse for images with high resolution and large contour density. In this essence we designed two parallel algorithms; Contour Parallelization Algorithm (CPA) and Partial Retrieval Parallelization Algorithm (PRPA). Both algorithms have considerably improved the performance of CBIS for both contour shape matching as well as partial image retrieval. To improve the effectiveness of CBIS in segmenting images with inhomogeneous backgrounds we used the phase congruency invariant features of Fourier transform components to highlight boundaries of objects prior to extracting their contours. The contour matching process has also been improved by constructing a fuzzy contour matching system that allows unbiased matching decisions. Further improvements have been achieved through the use of a contour tailored Fourier descriptor to make translation and rotation invariance. It is proved to be suitable for general contour shape matching where translation, rotation, and scaling invariance are required. For those images which are hard to be classified by object contours such as bacterial images, we define a multi-level cosine transform to extract their texture features for image classification. The low frequency Discrete Cosine Transform coefficients and Zenike moments derived from images are trained by Support Vector Machine (SVM) to generate multiple classifiers.

碩士
國立中興大學
應用數學系所
103
In many areas of imaging science, it is difficult to determine the phase of general measurements. For example, in the X-ray imaging, the detector can only measure the Fourier magnitude of the received optical wave. The classical phase retrieval is the problem of retrieving Fourier phase information from Fourier magnitude data. One tries to recover the unknown signal (or image) from measurements. Albert Fannjiang introduced randomly fabricated mask for the reconstruction of Fourier phase retrieval. The main purpose of the mask is to achieve the highly probable uniqueness of solutions up to a global phase factor. In this paper, we provide a number of experiments to explain the situation where traditional algorithms couldn’t successfully reconstruct unknown signals. In order to solve the Fourier phase retrieval, our contributions are the following: 1. We use combinations of algorithms to improve the performance of the traditional methods. 2. Among the proposed combinations of algorithms, the performance of nisvpTV2 algorithm combination is the best among the other combinations of algorithms.

碩士
國立中正大學
通訊工程學系
98
OFDM technology has been widely used in many wideband communication systems, for example, it has been applied in high speed personal area networks. OFDM system has many advantages, however, it has one major drawback at the transmitter, the high peak-to-average power ratio (PAPR). High PAPR increases the complexity of the ADC and DAC converters, and reduces the complexity of the transmitter power amplifier. We focus on exploring algorithm complexity of PTS system in this thesis. To fine the optimal solution of PTS algorithm, we need to perform exhaustive search with all combinations of rotation phases. However, the search complexity increases exponentially with the number of sub-blocks. In previous study, we have found some useful features of PTS-based OFDM systems. Several combinations of rotation phases can yield to the same PAPR output. Elegant relationships among these combinations help us to calculate them in advance. With these properties, the exhaustive search algorithm just needs to perform once for the set with same PAPR output combinations, and hence, dramatically reduces the search complexity. In this thesis, we apply the elegant features of PTS-bases OFDM systems to several sub-optimal search algorithms to reduce their search complexities. With computer simulations, we find that the search complexity for all these sub-optimal search algorithms has been significantly reduced under the same PAPR performance.

博士
國立交通大學
機械工程系所
102
This paper is aimed at developing a numerical method for two-phase flows with phase change on unstructured grids. In this article, two schemes are presented based on VOF (volume-of-fluid) method. The first scheme is to capture the interface by solving the advection equation of the volume fraction directly, termed as FBICS. In order to maintain the sharpness and boundedness of the interface, the convective flux through each cell face is determined by means of flux blending. The weakness of this method is that the interface region will occupy several grid spaces. In the other scheme (termed as CISIT), the interface is reconstructed first using interpolation practice, following by a predicted-correction procedure to handle the movement of the interface. Except for the interface cells, the VOF distribution is uniform, either in 1 or 0, and the interface occupies only one cell in its width. Unlike the PLIC method, the CISIT can be easily extended to unstructured grids with arbitrary geometry and 3-D problems without causing any further complication because its formulation is very simple. In PLIC, the reconstruction of the interface is not straightforward and the procedure to advance the interface is complicated because a large number of interface configurations (16 configurations for 2-D flows and 64 for 3-D flows) must be considered for determining of the flux across cell faces. Tests on a number of cases reveal that results via these two schemes in this study, which can be used on the unstructured grids, give good agreement with exact solutions or experimental data of free surface flows. In order to simulate the two-phase flow with phase change, the CISIT method is extended to include heat and mass transfer due to phase change. The mass transfer across the phase boundary is determined by taking into account the mass and energy jump conditions at the interface and added as a source term in the continuity equation. Then, the interface is treated as an internal boundary condition in the temperature flied. Finally, the energy equation is solved in an implicit way. Besides, this method is also extended to simulate the heat transfer of two-fluid flows without phase change based on the assumption of the continuity conditions of the temperature and heat flux instead of jump conditions at the interface. Application to film boiling flow on a horizontal plate at a state near the critical pressure shows that the boiling mode will be different at different superheat temperatures. According to different superheat tempera- tures, the boiling flows can be divided into five modes: single-bubble mode ( ), single/multiple bubble mode ( ), single-jet mode ( ), double-bubble mode ( ), and double-jet mode ( ). In the single-bubble mode, good agreement with semi-empirical correlations was obtained in terms of averaged Nusselt number. Furthermore, simulation of film boiling flow on a cylinder demonstrates that this method is applicable to boiling flow with complex geometry. Finally, the CISIT method with phase change is modified to calculate three-dimensional two-phase flows. Unlike two-dimensional flow, the interface is reconstructed with several non-coplanar triangular interfaces within the grid. First, this method was tested through computations of a number of two-fluid flows without phase change to validate the capability of tracking the interface in three- dimensional flows. In addition, this method was also applied to simulated film boiling flow on a horizontal plate. It can be shown that the space and time averaged Nusselt numbers obtained from the current simulations have good agreement with the semi-empirical correlations of Klimenko, especially for .Finally, the heat transfer model without phase change was used to simulate the molten tin droplet in oil and the octane inlet in water.

碩士
國立成功大學
電機工程學系碩博士班
97
To improve contouring accuracy, how to reduce tracking error and contour error is one of the most important issues in high performance machining. The aim of this thesis is to apply the Zero Phase Error Tracking Controller (ZPETC) to the contour following control problem on X-Y table and to improve contouring accuracy by reducing tracking error. To achieve this goal, a precise model of the controlled system for the zero phase error tracking controller is needed. Otherwise, the phase lag due to inaccurate modeling will lead to significant tracking error. In addition, the efficiency of the zero phase error tracking controller will be affected by the external disturbance, load variation and other nonlinear phenomenon. To deal with this problem, the Genetic Algorithm (GA) is employed to properly tune the parameters of the Zero phase tracking error controller in this thesis. Experimental results show that the proposed approach can significantly reduce tracking error and improve the contouring accuracy.

碩士
國立臺灣大學
應用力學研究所
103
The main application scenario of this thesis is using Michelson interferometer as the non-destructive testing method, which measures tiny deformation of objects by retrieving the phase of the interference fringes generated due to deformation. To retrieve the deformation induced phase, we measure the interference fringe first and then analyze the phase by a process called phase-unwrapping. Trying to take advantages of the rapid development of CCD/CMOS, which provides a platform to significantly increase the spatial modulation frequency, Fourier transform based phase-unwrapping algorithm was adopted in this thesis. Trying to remove the bottleneck associated with phase-unwrapping, Fourier spectrum of interference fringe obtained with pre-introduced spatial carrier frequency so as to distinguish the correct direction of object deformation, was adopted. This approach first proposed by Mitsuo Takeda in 1981 is similar to the Doppler interferometer that solves the directional ambiguity by providing a frequency shift. The only difference lies on either temporal or spatial frequency was pre-introduced. It is with this pre-introduced spatial frequency shift, retrieving the phase information by using only one interference fringe (intensity map) becomes feasible. More specifically, the above-mentioned approach circumvent the disadvantages associated with phase-shifting algorithms such as the 5,1 phase-shifting algorithm, etc. that require more than one image to analyze the phase information. It is to be noted that retrieve phase map from intensity map with a single intensity map not only saves valuable computation time but also provides us with a platform for dynamic measurement as high-speed camera can be used to record the time-varying interference fringes (intensity maps) first and then compute phase map after the deformation is completed. Furthermore, in dealing with problems related to valid or effective functional domain (domain with valid interference fringes) and regions with vastly different signal-to-noise ratios (SNR), Windowed Fourier Transform (WFT) first proposed by Qian Kemao in 2004 was also introduced in this thesis. For phase unwrapping, this study used Least-Squares method to get the information of measured object rapidly. It is to be noted that this method leads to the use of Fourier transform to solve a Poisson’s equation with Neumann boundary conditions. As Fourier transform algorithm was used in converting the intensity map to the phase map and then perform phase-unwrapping, these algorithms developed in this thesis provides us with an opportunity to adopt the many attempts over the last 50 years in speeding up the computation time associated with Fourier transform. Some of these methods include Fast Fourier Transform (FFT), Sparse Fast Fourier Transform (SFFT) and hardware solution such as Graphic Processing Unit (GPU), etc. All of which can then be integrated to develop an ultrafast phase analysis system, which can found applications potentials ranging from in-situ real-time optical field measurement, production-need driven automatic optical inspection (AOI), etc. This works completed throughout this research include setting the Michelson interferometer, integrating MATLAB and LabVIEW to transfer experimentally induced optical intensity map to computers for signal post-processing, etc. To verify the overall effectiveness of this system, this study analyzed the phase information by measuring the mirror and stainless steel deformation by using the Michelson interferometer set up. The results of unwrapped phase matched the object deformation, successfully validated the accuracy and the feasibility of integrating these algorithms in this thesis.

碩士
國立臺灣科技大學
電機工程系
106
In the context of electronic warfare, electronic counter-devices on military vehicles must achieve high sensitivity and high direction-finding accuracy for radio waves, and can effectively combat advanced radar equipment. Because of its low computational complexity and simple principle, the phase interferometer algorithm is often used in direction finding technology. Direction of the signal is estimated by the phase difference received between two antennas and uses interferometer algorithm to solve ambiguity. In the one-dimensional phase interferometer nonlinear antenna array system, If the baseline is lengthened, the accuracy of the angle of the estimated signal can be more accurate, but the probability of ambiguity will also increase. At this time, we have to choose the best trade-off to get lowest root mean square error (RMSE). We use mathematical derivation to propose an RMSE formula for algorithms. By using the RMSE formula, we can simulate and choose the optimal array ratio between the receivers in the AWGN channel. In addition, we choose three different algorithms according to the difference in the value of the angle of the direction. By deriving its RMSE formula and comparing its performance, we can use this RMSE formula to compare the advantages and disadvantages. Finally, the optimal array ratio of RMSE minimization is proposed by this RMSE formula.

Stereo imaging is routinely used in Simultaneous Localization and Mapping (SLAM) systems for the navigation and control of autonomous spacecraft proximity operations, advanced robotics, and robotic mapping and surveying applications. A key step (and generally the most computationally expensive step) in the generation of high fidelity geometric environment models from image data is the solution of the dense stereo correspondence problem. A novel method for solving the stereo correspondence problem to sub-pixel accuracy in the Fourier frequency domain by exploiting the Convolution Theorem is developed. The method is tailored to challenging aerospace applications by incorporation of correction factors for common error sources. Error-checking metrics verify correspondence matches to ensure high quality depth reconstructions are generated. The effect of geometric foreshortening caused by the baseline displacement of the cameras is modeled and corrected, drastically improving correspondence matching on highly off-normal surfaces. A metric for quantifying the strength of correspondence matches is developed and implemented to recognize and reject weak correspondences, and a separate cross-check verification provides a final defense against erroneous matches. The core components of this phase based dense stereo algorithm are implemented and optimized in the Compute Uni ed Device Architecture (CUDA) parallel computation environment onboard an NVIDIA Graphics Processing Unit (GPU). Accurate dense stereo correspondence matching is performed on stereo image pairs at a rate of nearly 10Hz.

碩士
國立中山大學
電機工程學系研究所
104
A phase-locked-loop frequency-locking algorithm implemented on an FPGA is presented for an ultrasonic spray coating system. The algorithm generates a sine wave for driving the ultrasonic transducer and ensures the frequency of the sine wave to automatically follow the transducer resonance (or anti-resonance) frequency, thereby maximizing the conversion efficiency from electrical power to vibration power. By the observation, the transducer''s impedance becomes purely resistive at its resonance (or anti-resonance). So a digital phase locked loop is employed to track the transducer resonance (or anti-resonance) by adjusting the driving frequency to where the driving voltage and current are in phase. The targeted driving frequencies range from 20 kHz to 80 kHz. The experiments performed on different transducers in the ultrasonic spray coating system confirm the effectiveness of the algorithm.

碩士
國立臺灣科技大學
資訊工程系
97
Owing to the demand of more efficient technology about robot vision applications, the researches on intelligent pattern detection and recognition have been rapidly grown in recent years. In addition to that an adaptive pattern classifier may affect recognition results, one of the most important characteristics of an intelligent robot vision system is to employ a quick and efficient pattern extraction method. To achieve such a good, in this thesis, we present an efficient two-phase labeling connected components algorithm based on a two-scan structure. In the labeling step, unlike conventional labeling algorithms using the same labeling operations for each object pixel for labeling or relation checking, our algorithm executes different labeling operations for each pixel depending on its location in a row of contiguous object pixels. Thus, we can reduce many unnecessary labeling operations and lessen the execution time. As to the label relation table, we propose a novel table structure similar to a circle to resolve label equivalence between provisional label sets. It not only can record all information as same as conventional label relation tables, but also can reduce 1/3 memory space at most than the other so. The implementation of this structure only requires two 1-D arrays, and we can easily record the relation between provisional labels and their corresponding representative labels by use of our record procedure. Experimental results reveal that the performance of our algorithm is superior to those of all conventional labeling algorithms for both ordinary and noisy images under the common sequential execution hardware.

博士
元智大學
電機工程學系甲組
107
The main purpose of this dissertation is to investigate the binary tree algorithm based three-phase power flow analysis technique for microgrid planning and design. First, the equivalent models of major components of a distribution system are derived. Secondly, a simple but more powerful three-phase power flow analysis algorithm for unbalanced radial distribution systems is proposed, because of the different requirements for analyzing extremely huge-scale network problems. Third, a Monte Carlo based three-phase power flow method is proposed to evaluate the impact of a variety of a hybrid energy generation system on the distribution system. The proposed Monte Carlo based power flow method is developed based on the proposed deterministic power flow method and a commercial software package OPTIMUS. Various distributed generations (DGs) are also combined in the proposed method. In this dissertation, the hybrid energy system in the campus is used as a sample case. Then, a comprehensive study of the hybrid energy microgrid system, in terms of steady-state voltage deviations, power flows, power losses, reverse power flows, short-circuit fault currents, maximum allowable DGs capacity and voltage unbalance factor, is performed to evaluate the impacts of the hybrid energy system on the utility power grid. Fourth, according to the analysis of the power distribution system in campus, the power flow algorithm is used to calculate the transmission line loss and transformer loss of the whole distribution system as the basis for the improvement of the subsequent energy use efficiency, and then propose the replacement priority to effectively reduce the unnecessary power loss. The electricity consumption of the campus is reduced to achieve the goal of energy saving and carbon reduction. Finally, a methodology for determining the proper installed capacity of an electric scooter charging station on the campus is developed. The three-phase power flow is used to evaluate the whole system voltage variations, and voltage unbalances. Then, the proper installed capacity of the charging station can be determined to ensure the service stability and power quality. The results of this dissertation are of value to promote microgrid planning and design.

碩士
國立中央大學
光電科學與工程學系
101
Stability improvement of a phase-detection-based surface-plasmon-resonance system is the topic in this thesis.By using phase-shifting interferometry with a random-shift tolerant algorithm replace five steps algorithm to solve Hysteresis effect and the other something would influence phase-shift to improve stability of system.Currently random-shift tolerant algorithm in this thesis, the mean error was 0.0019 waves and also to solve the problem that the triangle signal the five steps algorithm cann’t deal with and the phase-shifting interferometry with a random-shift tolerant algorithm cann’t deal with the Gaussion beam of laser.Using Twyman-Green interferometer to do the Gaussion beam interference experiment. The resulting mean error of phase-shift was 0.031 waves.In SPR(Surface Plasmon Resonnace) system,it would enter the water to measure stability of the system.In entering water condition, the stability was 0.044radians in 300-second period .It was also better than 0.065radians the result was calculated by five steps algorithm.To using the new algorithm in SPR system deal with the major problem that the triangle signal five steps algorithm cann’t do and improve the stability slightly.

碩士
國立臺灣大學
資訊工程學研究所
90
This thesis reports a study on a two-phase hierarchical data-clustering algorithm that operates based on gravity theory in physics. The design of the two-phase algorithm is aimed at achieving high clustering quality. The main purpose of the first phase of the algorithm is to identify those subclusters that are of spherical shapes. In this phase, each data instance is regarded as an object in a multidimensional vector space and a simulation is conducted to figure out how these objects influence each other and merge to form subclusters due to the gravity force. Then, those subclusters with spherical shapes are identified and an abstract model is employed to record the essential information of each spherical subcluster. In the second phase of the algorithm, a simulation is conducted to figure out how the spherical subclusters merge to form clusters at higher levels of the dendrogram due to the gravity force. In the study reported in this thesis, comprehensive experiments were conducted to figure out the optimal ranges of parameter settings and how the two-phase clustering algorithm compares with the existing hierarchical clustering algorithms with respect to clustering quality. Experimental results reveal that, if the parameters are properly set, then the two-phase clustering algorithm generally is more powerful in identifying clusters with arbitrary shapes than the existing hierarchical clustering algorithms.

碩士
國立交通大學
工業工程與管理學系
100
This research is concerned with the scheduling of a flexible job-shop problem (called FJSP), which involves two sub-decisions: (1) operation-to-machine assignment (2) operation sequencing for each machine. Most prior studies proposed meta-heuristic algorithms for the FJSP problem. This research proposes a new solution representation (called Sop), which is intended to model a particular sequence for all operations. Given a particular Sop, by some heuristic rules, we can obtain its two corresponding sub-decisions; which in turn represents a particular scheduling solution. Based on the Sop representation, this research adopts the existing architecture of genetic algorithm (GA) and develops a meta-heuristic algorithm (called GA_ Sop) to solve the FJSP problem.

碩士
國立交通大學
工業工程與管理學系
100
This research is concerned with the scheduling of a flexible job-shop problem (called FJSP), which involves two sub-decisions: (1) operation-to-machine assignment (2) operation sequencing for each machine. Most prior studies proposed meta-heuristic algorithms for the FJSP problem. This research proposes a new solution representation (called Sop), which is intended to model a particular sequence for all operations. Given a particular Sop, by a heuristic rule, we can obtain its two corresponding sub-decisions; which in turn represents a particular scheduling solution. Based on the Sop representation, this research adopts the existing architecture of particle swarm optimization (PSO) algorithm and develops a three phases meta-heuristic algorithm (called PSO_ Sop) to solve the FJSP problem.

We formulate Local Shape Descriptor selection for model-based object recognition in range data as an optimization problem and offer a platform that facilitates a solution. The goal of object recognition is to identify and localize objects of interest in an image. Recognition is often performed in three phases: point matching, where correspondences are established between points on the 3-D surfaces of the models and the range image; hypothesis generation, where rough alignments are found between the image and the visible models; and pose refinement, where the accuracy of the initial alignments is improved. The overall efficiency and reliability of a recognition system is highly influenced by the effectiveness of the point matching phase. Local Shape Descriptors are used for establishing point correspondences by way of encapsulating local shape, such that similarity between two descriptors indicates geometric similarity between their respective neighbourhoods. We present a generalized platform for constructing local shape descriptors that subsumes a large class of existing methods and allows for tuning descriptors to the geometry of specific models and to sensor characteristics. Our descriptors, termed as Variable-Dimensional Local Shape Descriptors, are constructed as multivariate observations of several local properties and are represented as histograms. The optimal set of properties, which maximizes the performance of a recognition system, depend on the geometry of the objects of interest and the noise characteristics of range image acquisition devices and is selected through pre-processing the models and sample training images. Experimental analysis confirms the superiority of optimized descriptors over generic ones in recognition tasks in LIDAR and dense stereo range images.
Thesis (Ph.D, Electrical & Computer Engineering) -- Queen's University, 2009-09-01 11:07:32.084