Event stream processing (ESP) has become increasingly important in modern applications. In this dissertation, I focus on providing a robust ESP solution by meeting three major research challenges regarding the robustness of ESP systems: (1) while event constraint of the input stream is available, applying such semantic information in the event processing; (2) handling event streams with out-of-order data arrival and (3) handling event streams with interval-based temporal semantics. The following are the three corresponding research tasks completed by the dissertation: Task I - Constraint-Aware Complex Event Pattern Detection over Streams. In this task, a framework for constraint-aware pattern detection over event streams is designed, which on the fly checks the query satisfiability / unsatisfiability using a lightweight reasoning mechanism and adjusts the processing strategy dynamically by producing early feedback, releasing unnecessary system resources and terminating corresponding pattern monitor. Task II - Complex Event Pattern Detection over Streams with Out-of-Order Data Arrival. In this task, a mechanism to address the problem of processing event queries specified over streams that may contain out-of-order data is studied, which provides new physical implementation strategies for the core stream algebra operators such as sequence scan, pattern construction and negation filtering. Task III - Complex Event Pattern Detection over Streams with Interval-Based Temporal Semantics. In this task, an expressive language to represent the required temporal patterns among streaming interval events is introduced and the corresponding temporal operator ISEQ is designed.

Pattern detection is a well-studied area of computer vision, but still current methods are unstable in images of poor quality. This thesis describes improvements over contemporary methods in the fast detection of unseen patterns in a large corpus of videos that vary tremendously in colour and texture definition, captured "in the wild" by mobile devices and surveillance cameras. We focus on three key areas of this broad subject; First, we identify consistency weaknesses in existing techniques of processing an image and it's horizontally reflected (mirror) image. This is important in police investigations where subjects change their appearance to try to avoid recognition, and we propose that invariance to horizontal reflection should be more widely considered in image description and recognition tasks too. We observe online Deep Learning system behaviours in this respect, and provide a comprehensive assessment of 10 popular low level feature detectors. Second, we develop simple and fast algorithms that combine to provide memory- and processing-efficient feature matching. These involve static scene elimination in the presence of noise and on-screen time indicators, a blur-sensitive feature detection that finds a greater number of corresponding features in images of varying sharpness, and a combinatorial texture and colour feature matching algorithm that matches features when either attribute may be poorly defined. A comprehensive evaluation is given, showing some improvements over existing feature correspondence methods. Finally, we study random decision forests for pattern detection. A new method of indexing patterns in video sequences is devised and evaluated. We automatically label positive and negative image training data, reducing a task of unsupervised learning to one of supervised learning, and devise a node split function that is invariant to mirror reflection and rotation through 90 degree angles. A high dimensional vote accumulator encodes the hypothesis support, yielding implicit back-projection for pattern detection.

Στη συγκεριμένη διπλωματική εργασία γίνεται μια σχοινοτενής μελέτη των τεχνικών αναγνώρισης νέων γεγονότων (ανωμαλιών ή outliers) σε ευρεία σετ δεδομένων. Το απαράιτητο θεωρητικό background που απαιτείται για την κατανόηση των τεχνικών παρέχεται ξεχωριστά προκειμένου να εξασφαλιστεί η συνοχή του κειμένου. Στο πρώτο κεφάλαιο γίνεται εισαγωγή στην έννοια και στις εφαρμογές του novelty detection, ενώ παρέχεται μια πρώτη κατηγοριοποίηση των τεχνικών αυτών. Στο Κεφάλαιο 2 αναλύονται οι στατιστικές προσεγγίσεις που έχουν προταθεί, τόσο οι παραμετρικές όσο και οι μη-παραμετρικές. Στα κεφάλαια 3 και 4 γίνεται μια εισαγωγή στα νευρωνικά δίκτυα και στα SVM προκείμένου να εξηγηθεί η χρήση τους στις εφαρμογές αναγνώρισης νέων γεγονότων ή ανωμαλιών (Κεφάλαιο 5). Ολοκληρώνοντας στη συγκεκριμένη διπλωματική εργασία στατιστικές προσεγγίσεις καθώς επίσης και τεχνικές βασιζόμενες σε νευρωνικά δίκτυα και SVM παρουσιάζονται με σαφήνεια, για την ανίχνευση νέων γεγονότων, ενώ η συγκριτική μελέτη τους παρέχει έναν συνοπτικό οδηγό-εργαλέιο που συνοψίζει τα πλεονεκτήματα και τα μειονεκτήματα των παρουσιαθέντων τεχνικών. / -

En raison du développement et de la disponibilité des appareils d'impression et de numérisation, le nombre de documents contrefaits augmente rapidement. En effet, les documents de valeur ainsi que les emballages de produits sont de plus en plus ciblés par des duplications non autorisées. Par conséquent, différents éléments de sécurité (hologrammes, encres, papiers) ont été proposés pour prévenir ces actions illégales. Dans cette thèse, nous nous concentrons sur les éléments de sécurité imprimés qui offrent un haut niveau de sécurité et qui possèdent une mise en œuvre et une intégration simple. Nous présentons comment générer de nouveaux éléments de sécurité qui visent à protéger les documents de valeur et les emballages contre des processus de duplication non autorisés. Ces éléments nous permettent en outre de stocker une grande quantité d'informations cachées.La caractéristique principale de ces éléments de sécurité est leur sensibilité au processus d'impression et de numérisation. Cette sensibilité est obtenue à l'aide de motifs texturés spécifiques. Ces motifs sont des images binaires qui possèdent une structure sensible aux processus d'impression, de numérisation et de copie. Nous définissons les critères spécifiques qui doivent être respectés lors du choix de ces motifs texturés. La quantité d'information encodée dans l'image augmente avec le nombre de motifs texturés utilisées.En complément, nous proposons dans ce mémoire d'améliorer la robustesse de la détection des modules, pour tous les codes graphiques, par l'utilisation d'une nouvelle mesure d'erreur quadratique moyenne pondérée. L'utilisation de cette nouvelle mesure nous a permis d'augmenter de façon significative le taux de reconnaissance des modules lorsqu'ils sont utilisés dans des codes à barres standard à haute densité. Enfin, nous étudions expérimentalement plusieurs phénomènes : le processus physique d'impression et de numérisation, la séparation du bruit du scanner de celui de l'imprimante et les changements de couleurs après processus d'impression et de numérisation. Nous concluons à partir de cette étude expérimentale, que le processus d'impression et de numérisation ne peut pas être modélisé comme un loi Gaussienne. Nous mettons en avant que ce processus n'est ni blanc ni ergodique au sens large. / Due to the development and availability of printing and scanning devices, the number of forged/counterfeited valuable documents and product packages is increasing. Therefore, different security elements (holograms, inks, papers) have been suggested to prevent these illegal actions. In this thesis, we focus on printed security elements that give access to a high security level with an easy implementation and integration. We present how to generate several novel security elements that aim to protect valuable documents and packaging against unauthorized copying process. Moreover, these security elements allow us to store a huge amount of hidden information.The main characteristic of these security elements is their sensitivity to the print-and-scan process. This sensitivity stems from the use of specific textured patterns. These patterns, which are binary images, have a structure that changes during the printing, scanning and copying processes. We define new specific criteria that ensures the chosen textured patterns to have the appropriate property. The amount of additional information encoded in the patterns increases with the number of patterns used.Additionally, we propose a new weighted mean squared error measure to improve the robustness of module detection for any high density barcodes. Thanks to this measure, the recognition rate of modules used in standard high density barcodes after print-and-scan process can be significantly increased. Finally, we experimentally study several effects: the physical print-and-scan process, separation of scanner noise from printer noise and changes of colors after print-and-scan process. We conclude, from these experimental results, that the print-and-scan process cannot be considered as being a Gaussian process. It has been also highlighted that this process is neither white nor ergodic in the wide sense.

<p>Events play an important role in many computer systems, from small reactive embedded applications to large distributed systems. Many applications react to events generated by a graphical user interface or by external sensors that monitor the system environment, and other systems use events for communication and synchronisation between independent subsystems. In some applications, however, individual event occurrences are not the main point of concern. Instead, the system should respond to certain event patterns, such as "the start button being pushed, followed by a temperature alarm within two seconds". One way to specify such event patterns is by means of an event algebra with operators for combining the simple events of a system into specifications of complex patterns.</p><p>This thesis presents an event algebra with two important characteristics. First, it complies with a number of algebraic laws, which shows that the algebra operators behave as expected. Second, any pattern represented by an expression in this algebra can be efficiently detected with bounded resources in terms of memory and time, which is particularly important when event pattern detection is used in embedded systems, where resource efficiency and predictability are crucial.</p><p>In addition to the formal algebra semantics and an efficient detection algorithm, the thesis describes how event pattern detection can be used in real-time systems without support from the underlying operating system, and presents schedulability theory for such systems. It also describes how the event algebra can be combined with a component model for embedded system, to support high level design of systems that react to event patterns.</p>

The goal of this thesis is the development of a fast and robust algorithm that is able to detect patterns in flow fields independent from their orientation and adequately visualize the results for a human user. This thesis is an interdisciplinary work in the field of vector field visualization and the field of pattern recognition. A vector field can be best imagined as an area or a volume containing a lot of arrows. The direction of the arrow describes the direction of a flow or force at the point where it starts and the length its velocity or strength. This builds a bridge to vector field visualization, because drawing these arrows is one of the fundamental techniques to illustrate a vector field. The main challenge of vector field visualization is to decide which of them should be drawn. If you do not draw enough arrows, you may miss the feature you are interested in. If you draw too many arrows, your image will be black all over. We assume that the user is interested in a certain feature of the vector field: a certain pattern. To prevent clutter and occlusion of the interesting parts, we first look for this pattern and then apply a visualization that emphasizes its occurrences. In general, the user wants to find all instances of the interesting pattern, no matter if they are smaller or bigger, weaker or stronger or oriented in some other direction than his reference input pattern. But looking for all these transformed versions would take far too long. That is why, we look for an algorithm that detects the occurrences of the pattern independent from these transformations. In the second part of this thesis, we work with moment invariants. Moments are the projections of a function to a function space basis. In order to compare the functions, it is sufficient to compare their moments. Normalization is the act of transforming a function into a predefined standard position. Moment invariants are characteristic numbers like fingerprints that are constructed from moments and do not change under certain transformations. They can be produced by normalization, because if all the functions are in one standard position, their prior position has no influence on their normalized moments. With this technique, we were able to solve the pattern detection task for 2D and 3D flow fields by mathematically proving the invariance of the moments with respect to translation, rotation, and scaling. In practical applications, this invariance is disturbed by the discretization. We applied our method to several analytic and real world data sets and showed that it works on discrete fields in a robust way.

Events play an important role in many computer systems, from small reactive embedded applications to large distributed systems. Many applications react to events generated by a graphical user interface or by external sensors that monitor the system environment, and other systems use events for communication and synchronisation between independent subsystems. In some applications, however, individual event occurrences are not the main point of concern. Instead, the system should respond to certain event patterns, such as "the start button being pushed, followed by a temperature alarm within two seconds". One way to specify such event patterns is by means of an event algebra with operators for combining the simple events of a system into specifications of complex patterns. This thesis presents an event algebra with two important characteristics. First, it complies with a number of algebraic laws, which shows that the algebra operators behave as expected. Second, any pattern represented by an expression in this algebra can be efficiently detected with bounded resources in terms of memory and time, which is particularly important when event pattern detection is used in embedded systems, where resource efficiency and predictability are crucial. In addition to the formal algebra semantics and an efficient detection algorithm, the thesis describes how event pattern detection can be used in real-time systems without support from the underlying operating system, and presents schedulability theory for such systems. It also describes how the event algebra can be combined with a component model for embedded system, to support high level design of systems that react to event patterns.

This report aims to explore a possible transparent alternative to the black box approach of machine learning in identifying a ship’s type from simple movement data, consisting of a set of coordinates with timestamps. This is achieved by an application that converts the set of coordinates to vectors and assigns them various traits, such as turn radius, speed and distance traveled, and then identifying the correlation between collections of different values of these traits, called granules, and different ship types. The results show a definite connection between certain kinds of granules and certain ship types and lay the foundation for building a more well defined syntax for ship identification.

This dissertation focuses on the topic of building group generalization by considering the detection of building patterns. Generalization is an important research field in cartography, which is part of map production and the basis for the derivation of multiple representation. As one of the most important features on map, buildings occupy large amount of map space and normally have complex shape and spatial distribution, which leads to that the generalization of buildings has long been an important and challenging task. For social, architectural and geographical reasons, the buildings were built with some special rules which forms different building patterns. Building patterns are crucial structures which should be carefully considered during graphical representation and generalization. Although people can effortlessly perceive these patterns, however, building patterns are not explicitly described in building datasets. Therefore, to better support the subsequent generalization process, it is important to automatically recognize building patterns. The objective of this dissertation is to develop effective methods to detect building patterns from building groups. Based on the identified patterns, some generalization methods are proposed to fulfill the task of building generalization. The main contribution of the dissertation is described as the following five aspects: (1) The terminology and concept of building pattern has been clearly explained; a detailed and relative complete typology of building patterns has been proposed by summarizing the previous researches as well as extending by the author; (2) A stroke-mesh based method has been developed to group buildings and detect different patterns from the building groups; (3) Through the analogy between line simplification and linear building group typification, a stroke simplification based typification method has been developed aiming at solving the generalization of building groups with linear patterns; (4) A mesh-based typification method has been developed for the generalization of the building groups with grid patterns; (5) A method of extracting hierarchical skeleton structures from discrete buildings have been proposed. The extracted hierarchical skeleton structures are regarded as the representations of the global shape of the entire region, which is used to control the generalization process. With the above methods, the building patterns are detected from the building groups and the generalization of building groups are executed based on the patterns. In addition, the thesis has also discussed the drawbacks of the methods and gave the potential solutions.:Abstract I Kurzfassung III Contents V List of Figures IX List of Tables XIII List of Abbreviations XIV Chapter 1 Introduction 1 1.1 Background and motivation 1 1.1.1 Cartographic generalization 1 1.1.2 Urban building and building patterns 1 1.1.3 Building generalization 3 1.1.4 Hierarchical property in geographical objects 3 1.2 Research objectives 4 1.3 Study area 5 1.4 Thesis structure 6 Chapter 2 State of the Art 8 2.1 Operators for building generalization 8 2.1.1 Selection 9 2.1.2 Aggregation 9 2.1.3 Simplification 10 2.1.4 Displacement 10 2.2 Researches of building grouping and pattern detection 11 2.2.1 Building grouping 11 2.2.2 Pattern detection 12 2.2.3 Problem analysis . 14 2.3 Researches of building typification 14 2.3.1 Global typification 15 2.3.2 Local typification 15 2.3.3 Comparison analysis 16 2.3.4 Problem analysis 17 2.4 Summary 17 Chapter 3 Using stroke and mesh to recognize building group patterns 18 3.1 Abstract 19 3.2 Introduction 19 3.3 Literature review 20 3.4 Building pattern typology and study area 22 3.4.1 Building pattern typology 22 3.4.2 Study area 24 3.5 Methodology 25 3.5.1 Generating and refining proximity graph 25 3.5.2 Generating stroke and mesh 29 3.5.3 Building pattern recognition 31 3.6 Experiments 33 3.6.1 Data derivation and test framework 33 3.6.2 Pattern recognition results 35 3.6.3 Evaluation 39 3.7 Discussion 40 3.7.1 Adaptation of parameters 40 3.7.2 Ambiguity of building patterns 44 3.7.3 Advantage and Limitation 45 3.8 Conclusion 46 Chapter 4 A typification method for linear building groups based on stroke simplification 47 4.1 Abstract 48 4.2 Introduction 48 4.3 Detection of linear building groups 50 4.3.1 Stroke-based detection method 50 4.3.2 Distinguishing collinear and curvilinear patterns 53 4.4 Typification method 55 4.4.1 Analogy of building typification and line simplification 55 4.4.2 Stroke generation 56 4.4.3 Stroke simplification 57 4.5 Representation of newly typified buildings 60 4.6 Experiment 63 4.6.1 Linear building group detection 63 4.6.2 Typification results 65 4.7 Discussion 66 4.7.1 Comparison of reallocating remained nodes 66 4.7.2 Comparison with classic line simplification method 67 4.7.3 Advantage 69 4.7.4 Further improvement 71 4.8 Conclusion 71 Chapter 5 A mesh-based typification method for building groups with grid patterns 73 5.1 Abstract 74 5.2 Introduction 74 5.3 Related work 75 5.4 Methodology of mesh-based typification 78 5.4.1 Grid pattern classification 78 5.4.2 Mesh generation 79 5.4.3 Triangular mesh elimination 80 5.4.4 Number and positioning of typified buildings 82 5.4.5 Representation of typified buildings 83 5.4.6 Resizing Newly Typified Buildings 85 5.5 Experiments 86 5.5.1 Data derivation 86 5.5.2 Typification results and evaluation 87 5.5.3 Comparison with official map 91 5.6 Discussion 92 5.6.1 Advantages 92 5.6.2 Further improvements 93 5.7 Conclusion 94 Chapter 6 Hierarchical extraction of skeleton structures from discrete buildings 95 6.1 Abstract 96 6.2 Introduction 96 6.3 Related work 97 6.4 Study area 99 6.5 Hierarchical extraction of skeleton structures 100 6.5.1 Proximity Graph Network (PGN) of buildings 100 6.5.2 Centrality analysis of proximity graph network 103 6.5.3 Hierarchical skeleton structures of buildings 108 6.6 Generalization application 111 6.7 Experiment and discussion 114 6.7.1 Data statement 114 6.7.2 Experimental results 115 6.7.3 Discussion 118 6.8 Conclusions 120 Chapter 7 Discussion 121 7.1 Revisiting the research problems 121 7.2 Evaluation of the presented methodology 123 7.2.1 Strengths 123 7.2.2 Limitations 125 Chapter 8 Conclusions 127 8.1 Main contributions 127 8.2 Outlook 128 8.3 Final thoughts 131 Bibliography 132 Acknowledgements 142 Publications 143

The goal of this thesis is the development of a fast and robust algorithm that is able to detect patterns in flow fields independent from their orientation and adequately visualize the results for a human user. This thesis is an interdisciplinary work in the field of vector field visualization and the field of pattern recognition. A vector field can be best imagined as an area or a volume containing a lot of arrows. The direction of the arrow describes the direction of a flow or force at the point where it starts and the length its velocity or strength. This builds a bridge to vector field visualization, because drawing these arrows is one of the fundamental techniques to illustrate a vector field. The main challenge of vector field visualization is to decide which of them should be drawn. If you do not draw enough arrows, you may miss the feature you are interested in. If you draw too many arrows, your image will be black all over. We assume that the user is interested in a certain feature of the vector field: a certain pattern. To prevent clutter and occlusion of the interesting parts, we first look for this pattern and then apply a visualization that emphasizes its occurrences. In general, the user wants to find all instances of the interesting pattern, no matter if they are smaller or bigger, weaker or stronger or oriented in some other direction than his reference input pattern. But looking for all these transformed versions would take far too long. That is why, we look for an algorithm that detects the occurrences of the pattern independent from these transformations. In the second part of this thesis, we work with moment invariants. Moments are the projections of a function to a function space basis. In order to compare the functions, it is sufficient to compare their moments. Normalization is the act of transforming a function into a predefined standard position. Moment invariants are characteristic numbers like fingerprints that are constructed from moments and do not change under certain transformations. They can be produced by normalization, because if all the functions are in one standard position, their prior position has no influence on their normalized moments. With this technique, we were able to solve the pattern detection task for 2D and 3D flow fields by mathematically proving the invariance of the moments with respect to translation, rotation, and scaling. In practical applications, this invariance is disturbed by the discretization. We applied our method to several analytic and real world data sets and showed that it works on discrete fields in a robust way.