<p>This thesis covers topics relevant to model-based coding. Model-based coding is a promising very low bit rate video coding technique. The idea behind this technique is to parameterize a talking head and to extract and transmit the parameters describing facial movements. At the receiver, the parameters are used to reconstruct the talking head. Since only high-level animation parameters are transmitted, very high compression can be achieved with this coding scheme. This thesis covers the following three key problems.</p><p>Although it is a fundamental problem, the initialization problem, has been neglected some extent in the literature. In this thesis, we pay particular attention to the study of this problem. We propose a pseudo-automatic initialization scheme: an Analysis-by-Synthesis scheme based on Simulated Annealing. It has been proved to be an efficient scheme.</p><p>Owing to technical advance today and the newly emerged MPEG-4 standard, new schemes of performing texture mapping and motion estimation are suggested which use sample based direct texture mapping; the feasibility of using active motion estimation is explored which proves to be able to give more than 10 times tracking resolution. Based on the matured face detection technique, Dynamic Programming is introduced to face detection module and work for face tracking.</p><p>Another important problem addressed in this thesis is how to evaluate the face tracking techniques. We studied the evaluation problems by examining the commonly used method, which employs a physical magnetic sensor to provide "ground truth". In this thesis we point out that it is quite misleading to use such a method.</p>

This thesis covers topics relevant to model-based coding. Model-based coding is a promising very low bit rate video coding technique. The idea behind this technique is to parameterize a talking head and to extract and transmit the parameters describing facial movements. At the receiver, the parameters are used to reconstruct the talking head. Since only high-level animation parameters are transmitted, very high compression can be achieved with this coding scheme. This thesis covers the following three key problems. Although it is a fundamental problem, the initialization problem, has been neglected some extent in the literature. In this thesis, we pay particular attention to the study of this problem. We propose a pseudo-automatic initialization scheme: an Analysis-by-Synthesis scheme based on Simulated Annealing. It has been proved to be an efficient scheme. Owing to technical advance today and the newly emerged MPEG-4 standard, new schemes of performing texture mapping and motion estimation are suggested which use sample based direct texture mapping; the feasibility of using active motion estimation is explored which proves to be able to give more than 10 times tracking resolution. Based on the matured face detection technique, Dynamic Programming is introduced to face detection module and work for face tracking. Another important problem addressed in this thesis is how to evaluate the face tracking techniques. We studied the evaluation problems by examining the commonly used method, which employs a physical magnetic sensor to provide "ground truth". In this thesis we point out that it is quite misleading to use such a method.

Nachrichtentechnische Systeme werden stets durch unvermeidbare zufällige Störungen beeinflußt. Neben anderen Komponenten sind davon besonders Oszillatoren betroffen. Die durch die Störungen verursachten zufälligen Schwankungen in der Oszillatorausgabe können als Amplituden- und Phasenabweichungen modelliert werden. Dabei zeigt sich, daß vor allem zufällige Phasenfluktuationen von Bedeutung sind. Zufällige Phasenfluktuationen können unter Verwendung stochastischer Prozesse zweiter Ordnung mit kurzem oder langem Gedächtnis modelliert werden. Inhalt der Dissertation ist die Herleitung eines Verfahrens zur Simulation zufälliger Phasenfluktuationen von Oszillatoren mit kurzem Gedächtnis unter Berücksichtigung von Datenblattangaben.

The objective of this research is to demonstrate how an underlying system's state vector distribution can be determined in a distributed heterogeneous sensor network with reduced subspace observability at the individual nodes. We show how the network, as a whole, is capable of observing the target state vector even if the individual nodes are not capable of observing it locally. The initialization algorithm presented in this work can generate the initial state vector distribution for networks with a variety of sensor types as long as the measurements at the individual nodes are known functions of the target state vector. Initialization is accomplished through a novel distributed implementation of the particle filter that involves serial particle proposal and weighting strategies, which can be accomplished without sharing raw data between individual nodes in the network. The algorithm is capable of handling missed detections and clutter as well as compensating for delays introduced by processing, communication and finite signal propagation velocities. If multiple events of interest occur, their individual states can be initialized simultaneously without requiring explicit data association across nodes. The resulting distributions can be used to initialize a variety of distributed joint tracking algorithms. In such applications, the initialization algorithm can initialize additional target tracks as targets come and go during the operation of the system with multiple targets under track.

Kvantová výpočetní technika je rychle rostoucí obor informatiky, který přenáší principy kvantových jevu do našeho každodenního života. Díky své kvantové podstatě získávají kvantové počítače převahu nad klasickými počítači. V této práci jsme se zaměřili na vysvětlení základů kvantového počítání a jeho implementaci na kvantovém počítači. Zejména se zaměřujeme na popis fungování, konstrukci a implementaci Groverova algoritmu jako jednoho ze základních kvantových algoritmů. Demonstrovali jsme sílu tohoto kvantového algoritmu při prohledávání databáze a porovnávali ho s klasickými nekvantovými algoritmy pomocí implementace prostřednictvím simulačního prostředí QISKit. Pro simulaci jsme použili QASM Simulator a State vector Simulator Aer backends a ukázali, že získané výsledky korelují s dříve diskutovanými teoretickými poznatky. Toto ukazuje, že Groverův algoritmus umožňuje kvadratické zrychlení oproti klasickému nekvantovému vyhledávacímu algoritmu, Použitelnost algoritmu stejně jako ostatních kvantových algoritmů je ale stále omezena několika faktory, mezi které patří vysoké úrovně dekoherence a chyby hradla.

This master's thesis deals with open source PBX security against security attacks. In the theoretical part is detailed description of problematic about attacks that could be used on VoIP systems with high focus on the Denial of Service attack. Furthermore are in theoretical part described methods of security of initialization protocol SIP. Individual chapter is devoted to intrusion detection and prevention of IDS and IPS systems, focusing on Snort and OSSEC. In the practical part of the work was created generator of attacks against various PBX systems, which was subsequently used for detailed testing. Special tests of PBX system are then used against DoS attacks, for which was created protection in form of active elements consisting of IDS Snort & OSSEC. These are capable to provide protection in real-time. The protection was tested on particular PBX systems and in matter of comparison were measured possibilities before and after of security implementation. The output of this work is attacks generator VoIPtester and creation of configuration rules for Snort and OSSEC.

Deep neural networks (DNN) have been widely demonstrated to be more powerful than their shallower counterparts in a variety of computer vision tasks and remote sensing applications. However, as many techniques are based on trial-and-error experiments as opposed to systematic evaluation, scientific evidence for the superiority of DNN needs more theoretical and experimental foundations. Recent work has shown that the neural network expressivity, measured by the number of linear regions, is independent of the network structure, suggesting that the success of deep neural networks is attributed to its ease of training. Inspired by this, this project aims to investigate novel approaches to train neural networks and obtain desired properties of the regional properties of linear regions. In particular, it highlights the regional structure of linear regions in different decision regions and seeks to initialize the network in a better position that makes it easier to have this regional structure. By counting the total number of linear regions in the input space, we validated that the shallow wider networks and the deep narrow networks share the same upper-bound expressivity in different synthetic datasets. We also discovered that the linear regions along the decision boundary are larger in shape and fewer in number, while being denser and fitted to the data manifold when close to the data. Our experiments indicate that the proposed initialization method can generate more linear regions at initialization, make the training converge faster, and finally generate linear regions that better fit the data manifold on synthetic data. On the EuroSAT satellite dataset, the proposed initialization method does not facilitate the convergence of ResNet-18, but achieves better performance with an average increase of 0.14% on accuracy compared to pre-trained weights and 0.19% compared to He uniform initialization. / Djupa neurala nätverk (Deep Neural Networks, DNN) har i stor utsträckning visat sig vara mer kraftfulla än sina grunda motsvarigheter i en mängd olika datorseendeuppgifter och fjärranalystillämpningar. Många tekniker är dock baserade på försök och misstag snarare än systematisk utvärdering, och vetenskapliga bevis för DNN:s överlägsenhet behöver mer teoretiska och experimentella grunder. Nyligen utförda arbeten har visat att det neurala nätverkets uttrycksförmåga, mätt som antalet linjära regioner, är oberoende av nätverksstrukturen, vilket tyder på att framgången för djupa neurala nätverk beror på att de är lätta att träna. Inspirerat av detta syftar detta projekt till att undersöka nya metoder för att träna neurala nätverk och få önskade egenskaper hos de regionala egenskaperna hos linjära regioner. I synnerhet belyser det den regionala strukturen hos linjära regioner i olika beslutsregioner och försöker initiera nätverket i ett bättre läge som gör det lättare att få denna regionala struktur. Genom att räkna det totala antalet linjära regioner i ingångsutrymmet validerade vi att de grunda bredare nätverken och de djupa smala nätverken har samma övre gräns för uttrycklighet i olika syntetiska dataset. Vi upptäckte också att de linjära regionerna längs beslutsgränsen är större till formen och färre till antalet, samtidigt som de är tätare och anpassade till datamångfalden när de ligger nära data. Våra experiment visar att den föreslagna initialiseringsmetoden kan generera fler linjära regioner vid initialiseringen, få träningen att konvergera snabbare och slutligen generera linjära regioner som bättre passar datamångfalden på syntetiska data. På EuroSAT-satellitdatamängden underlättar den föreslagna initialiseringsmetoden inte konvergensen för ResNet-18, men uppnår bättre prestanda med en genomsnittlig ökning av noggrannheten med 0,14% jämfört med förtränade vikter och 0,19% jämfört med He uniform initialisering.

Clustering of observations into groups arises as a fundamental challenge both in academia and industry. Many clustering algorithms exist, and the most widely used clustering algorithm, the K-means, notably suffers from sensitivity to initial allocation of cluster centers. Moreover, many heuristics and algorithms have been developed to find the best initial allocation, and this experimental study compares methods of initialization by measuring how well the initialization methods perform on simulated, small datasets, through various performance criterion. The results show that using the output clusters of a Hierarchical clustering is the best initialization method. Moreover, the most popular methods, Random partitioning and KMeans++, perform poorly. Although the experimental setup may favour some initialization methods over others, the applied researchers are recommended to perform a Hierarchical clustering as an initialization of the K-means algorithm.

In modern combat aircraft, the most common localization method integrates a Global Navigation Satellite System (GNSS) with an Inertial Navigation System (INS). Although GNSS is the optimal choice for navigation, there are situations when the GNSS satellite signal is unavailable. This can happen due to various reasons such as jamming, physical obstacles, or technical failures. An alternative method to GNSS is utilizing Signals of Opportunity (SOP), which leverages signals not intended for navigation, such as those from cellular towers. These signals are transmitted from non-controllable sources, and challenges may arise due to the lack of guarantee regarding their quality and availability. Therefore, it is crucial that any estimation method utilizing SOP is robust to ensure accurate aircraft localization. This thesis investigates three different localization approaches to address this challenge. This study explores SOP sources with both known and unknown positions. For known signal source positions, an Extended Kalman Filter (EKF) based solution is utilized as a baseline to evaluate how well unknown signal sources can be used to estimate the aircraft's location. To address the challenge of unknown signal source positions, an EKF combined with a Simultaneous Localization and Mapping (SLAM) method, referred to as EKF SLAM, is used. In this case, the sources are introduced through two different approaches. The first approach, undelayed initialization, introduces the signal source directly when observed. The second approach, delayed initialization, involves inverse depth parameterization (IDP) and preprocessing of the signal source position before fully introducing it into the aircraft system. While both approaches outperform an unassisted INS approach, they do not achieve the same level of performance as when the source positions are known. Moreover, various factors, including the aircraft's trajectory, measurement noise, measurement frequency, and the initial covariance of new landmarks, influence the performance of the EKF SLAM approaches. Additionally, delayed initialization is strongly influenced by a threshold assessing landmark position estimate linearity, underscoring its sensitivity to accuracy. The concept behind delayed initialization aims to reduce the error of the signal source position before it is introduced to the system. This method has been proven to significantly reduce the signal source position error. However, its robustness is influenced by several factors, including the parallax angle, sudden changes in the aircraft's direction, and particularly the initial covariance of a landmark estimate. The accuracy of the aircraft's position is crucial, resulting in a trade-off between preprocessing and rapidly initializing a signal source position to the aircraft system. In contrast, undelayed initialization is less sensitive to trajectory changes, even though it introduces the signal sources with greater initial error. There is a significant difference in computational time when comparing known and unknown sources. As the number of sources increases, the computational time for unknown sources is more affected than for known sources. The delayed source initialization method increases computational time due to its preprocessing, especially as more sources are used. Conversely, initializing sources directly reduces the computational time, as no preprocessing is required. / I moderna stridsflygplan är den vanligaste lokaliseringsmetoden att integrera ett Global Navigation Satellite System (GNSS) med ett Inertial Navigation System (INS). Även om GNSS är det optimala valet för navigation finns det situationer när GNSS-satellitsignalen inte är tillgänglig. Detta kan inträffa på grund av olika orsaker som störningar, fysiska hinder eller tekniska fel. En alternativ metod till GNSS är att använda Signals of Opportunity (SOP), som utnyttjar signaler som inte är avsedda för navigation, till exempel de från mobilmaster. Dessa signaler kommer från okontrollerbara källor, vilket kan medföra utmaningar på grund av att deras kvalitet och tillgänglighet inte kan garanteras. Därför är det viktigt att varje lokaliseringsmetod som använder SOP är robust för att säkerställa en bra och korrekt flygplans positionering. Detta examensarbete undersöker tre olika lokaliseringsmetoder för att hantera denna utmaning. Denna studie utforskar SOP-källor med både kända och okända positioner. För kända positioner används en lösning baserad på ett Extended Kalman Filter (EKF) som en baslinje för att utvärdera hur väl okända signalkällor kan användas för att uppskatta flygplanets position. För att hantera utmaningen med okända signalkällors positioner används ett EKF kombinerad med en metod vid namn Simultaneous Localization and Mapping (SLAM), även kallad EKF SLAM. I detta fall introduceras källorna genom två olika tillvägagångssätt. Det första tillvägagångssättet, ofördröjd initialisering, introducerar signalkällan direkt när den observeras. Det andra tillvägagångssättet, fördröjd initialisering, involverar inverse depth parameterization (IDP) och förbearbetning av signalkällans position innan den introduceras i flygplanets lokaliseringssystem. Även om båda tillvägagångssätten presterar bättre än en oassisterad INS-metod uppnår de inte samma prestandanivå som när källornas position är kända. Dessutom påverkar olika faktorer prestandan hos EKF SLAM-metoderna, vilka främst är flygplanets flygbana, mätbrus, mätfrekvens och den initiala kovariansen av nya landmärken. Dessutom påverkas fördröjd initialisering starkt av en tröskel som bedömer linjäritet hos landmärkes positionen, vilket understryker dess känslighet för noggrannhet. Konceptet bakom fördröjd initialisering syftar till att minska felet i signalkällans position innan den introduceras i lokaliseringssystemet. Denna metod har visat sig kunna minska felet i signalkällans position avsevärt. Emellertid påverkas dess robusthet av flera faktorer, inklusive parallaxvinkeln, plötsliga förändringar i flygplanets riktning och särskilt den initiala kovariansen av uppskattningen av ett landmärkes position. Noggrannheten i flygplanets position är avgörande, vilket resulterar i en avvägning mellan förbearbetning och snabb initialisering av en signalkällas position till flygplanets lokaliseringssystem. Till skillnad från fördröjd initialisering är ofördröjd initialisering mindre känslig för förändringar i flygbanan, även om den introducerar signalkällorna med större initialt fel. Det finns en anmärkningsvärd skillnad i beräkningstid när man jämför kända och okända källors. När antalet källor ökar påverkas beräkningstiden för okända källor mer än för kända källor. Den fördröjda källinitialiseringsmetoden ökar beräkningstiden på grund av dess förbearbetning, särskilt när många källor används. Däremot minskar beräkningstiden när källor initialiseras direkt, eftersom ingen förbearbetning krävs.

[pt] Esta Dissertação procura, sob um enfoque freqüentista, discutir tecnologias para que se imponham restrições no processo de estimação de componentes não observáveis associadas a um modelo em Espaço de Estado (EE) arbitrário. O escopo do texto abrange desde procedimentos propostos pioneiramente por Howard Doran para restrições de igualdade, lineares e/ou não lineares, invariantes ou variantes no tempo, em modelos em EE lineares, até a adoção e o ajuste de estruturas mais delicadas, como os modelos em EE não lineares. Entende-se que estes últimos se constituem em uma alternativa relevante, caso seja requerida, por exemplo, a imposição de restrições de desigualdade. Técnicas e estratégias de implementação são apresentadas, debatidas e comparadas, incluindo-se também o processo de estimação de parâmetros desconhecidos e a questão de diagnósticos. Ao final, são apresentados exercícios empíricos com base nas tecnologias discutidas. Os modelos propostos para esta ilustração visam à realização da análise dinâmica de estilo baseado no retorno para carteiras de investimento brasileiras (a versão estática desses modelos fora introduzida por William Sharpe, para carteiras norte-americanas), os quais devem, eventualmente, abranger dois tipos de restrições nas componentes de interesse, quais sejam, um de igualdade e outro de desigualdade. / [en] This Dissertation aims, in a frequentist way, to discuss technologies for imposing restrictions in non-observable components associated with an arbitrary State Space (SS) model. The text scope ranges from procedures proposed originally by Howard Doran for equality, linear or non- linear, time invariant or time varying restrictions in a linear SS model, to adoption and estimation of more complicated structures like non-linear SS models. It is understood that these last ones are a relevant alternative, in cases of, for instance, inequality restrictions requirement. Implementation techniques and strategies are given, debated and compared, also including unknown parameters estimation and diagnostics analysis. At the end, empirical exercises are presented based on discussed methodologies. The proposed models for this illustration aim at dynamic return based style analysis for Brazilian investment portfolios (the static version of these models had been introduced by William Sharpe, for American portfolios), which shall eventually satisfy two kinds of restrictions on components of interest, namely one of equality and other of inequality.