No description available.

Thesis (M.Phil.)--Hong Kong University of Science and Technology, 2008. / Includes bibliographical references (leaves 82-85). Also available in electronic version.

From 1927 through 1939, Salvador Dalí went through an arduous artistic pursuit to visualize his perspective on reality. Dalí’s inter-connected visual and verbal process lasted for over a decade, during which he went from confronting reality to accepting that the world is an irrational paradox. This study asserts that his investigation took him from the fragmented images of cinema to the metamorphic shapes of the ‘soft and hard’ and ultimately to a series of multiple images, which envisioned his paranoid-critical method. In his 1930 article “The Rotting Donkey,” Dalí wrote, “I believe that the moment is near when, through a process of thought of a paranoiac and active character, it will be possible to systematize confusion and contribute to the total discrediting of the world of reality.” By 1930, Dalí knew he wanted to completely “discredit” the world and he saw Albert Einstein’s space-time and Sigmund Freud’s discovery of the unconscious as proof for the existence of a new dimension of experience. In 1933, Dalí found further evidence for his viewpoint in Jacques Lacan’s writings on paranoia. While Dalí wrote of and incorporated Einstein’s, Freud’s and Lacan’s ideas into his art, he came to his own conclusions on reality, which he gave form to in his artworks. To illuminate Dalí’s creative process this thesis will examine Dalí’s infamous film Un chien andalou (1929), his iconic painting The Persistence of Memory (1931), and his under-appreciated masterpiece The Endless Enigma (1938), as he visually expressed his changes in thought most clearly and convincingly in these three works.

Bedeutung im Kontext von Schäden erweitert. Es wird eine systematische Methodik für die Schadensidentifikation entwickelt, die die Datenerfassung, Schadensbewertung und Schadensdatenverwaltung umfasst. Für die Erfassung von sichtbaren Schadensdaten wird die semantische Segmentierung verwendet. Zahlreiche Strategien und innovative Convolutional Neural Networks wurden entwickelt, um herkömmliche Netzwerke im Kontext der semantischen Segmentierung zu verbessern. Allerdings wurde ein umfassender Vergleich dieser Netzwerke selten durchgeführt. Für zwei Strategien, der Attention Mechanismen und der Generative Adversarial Networks, wird eine vergleichende Studie durchgeführt, um die semantische Segmentierung zu verbessern. Basierend auf dem U-Net werden neuartige Verteilungstypen für beide Strategien mit verschiedenen Diskriminatoren entwickelt und verglichen. Die am besten abschneidenden Netzwerke werden dann einem Validierungsprozess unterzogen, und auch die kombinierten Effekte der beiden Strategien werden vertieft untersucht. Die Cloud Simulation wird zur numerischen Bewertung von Brückenschäden und der Identifikation von verdeckten Schäden angewendet. Zwei Ansätze, nämlich der Single Variation Approach (SVA) und der Dual Variation Approach (DVA), werden vorgestellt. Beide Ansätze werden auf unterschiedliche Szenarien angewendet, um den Einfluss verschiedener Lastfälle und Überwachungspunkte zu studieren. Der effektivere DVA-Ansatz wird in einem Prototyp implementiert, der Funktionalitäten wie Datenkonvertierung, Visualisierung, Generierung von Modellvariationen und Ergebnisanalyse umfasst. Zur Validierung wird eine Stahlbetonbrücke analysiert. Qualitative und quantitative Bewertungen für die Schadensrehabilitation werden in eine Wissensbasis integriert, die automatische Vorschläge für die praktische Schadensrehabilitation für die infizierten Schäden liefert. Augmented Reality wird zur Verbesserung des Visualisierungsergebnisses bei der Schadensinformationsverwaltung für die Vor-Ort-Inspektion und die Rehabilitationsinformationen eingesetzt und in einer Baustellenumgebung validiert. Den Abschluss bildet eine Marketingperspektive der neugewonnenen Ergebnisse. / The entire research work is dedicated to reinterpreting the concept of system identification by exploring its connotation and expanding its denotation in the context of damage. The work focuses on bridges as representative structures and proposes a systematic methodology for damage identification, encompassing damage data acquisition, damage assessment, and damage data management.  Semantic segmentation is employed for viewable damage data identification. Numerous strategies and innovative convolutional neural networks have been developed to enhance traditional networks in the context of semantic segmentation. However, a comprehensive comparison of these networks has been rarely conducted. Two strategies, namely attention mechanisms and generative adversarial networks, are examined in order to enhance semantic segmentation. Based on the U-net, novel distribution types of attention mechanisms and generative adversarial networks with different discriminators are compared in a lightweight test. The best performed networks are then implemented in the validation process, and in addition the combined effects of the attention mechanism and discriminator are investigated. Cloud simulation is applied for quantitative evaluation and identification of non-viewable damage. Two approaches, namely the Single Variation Approach (SVA) and the Dual Variation Approach (DVA), are introduced and applied to different scenarios to account for various load cases and monitoring points as variables. A prototype is developed to implement the more effective DVA approach, incorporating functionalities such as data conversion, visualization, model variation generation and result analysis. A monitored concrete bridge is employed for validation of the assessment of the effectiveness and reliability of the method. Qualitative and quantitative assessments are incorporated into a knowledge base for damage rehabilitation, which automatically provide practical suggestion for the specific identified damage.  Augmented reality is utilized to enhance the visualization experience for on-site inspection providing rehabilitation information and a prototype in a construction setting. The conclusion presents a marketing perspective on the findings.

This thesis deals with the issue of creating applications of augmented reality. There are described types of augmented reality, problems associated with developing of applications of this kind. Also there is description of method for detection of moving objects from the record of stationary camera. The thesis deals with camera calibration using knowledge of scene geometry. Scene editor for scene geometry estimation is included in this work. Augmented reality is represented by physical simulations, where object from virtual reality interacts with detected objects and with objects defined in scene editor.

Visual Simultaneous Localisation and Mapping (SLAM) is concerned with simultaneously estimating the pose of a camera and a map of the environment from a sequence of images. Traditionally, sparse maps comprising isolated point features have been employed, which facilitate robust localisation but are not well suited to advanced applications. In this thesis, we present map representations that allow a more dense description of the environment. In one approach, planar features are used to represent textured planar surfaces in the scene. This model is applied within a visual SLAM framework based on the Extended Kalman Filter. We presents solutions to several challenges which arise from this approach.