Tesis sobre el tema "Caméra 360"

Dans les environnements professionnels contemporains, le travail est souvent dispersé sur différents lieux géographiquement éloignés. La réunion des collaborateurs en présentiel pouvant s'avérer complexe, les professionnels se reposent aujourd'hui sur les technologies de l'information et de la communication pour organiser les interactions. La télé-immersion s'inscrit dans cette continuité de technologies avec l'ambition de rapprocher les individus séparés géographiquement comme s'ils étaient présents dans un même lieu. Cette thèse propose de réaliser un système de télé-immersion original basé sur une caméra 360° et des casques de réalité étendue. Ce système est conçu pour que des utilisateurs dans un lieu d'intérêt puissent ramener auprès d'eux des utilisateurs distants simplement en posant une caméra 360°. Grâce à la réalité étendue, les utilisateurs distants sont comme téléportés sur le lieu d'intérêt. Le système est spécifiquement développé pour répondre au besoin de nouvelles technologies d'enseignement à distance, afin que des enseignants puissent dispenser des cours immersifs à des étudiants chez eux.Le premier verrou pour atteindre un tel système consiste à gérer le point de vue de multiples utilisateurs avec une seule caméra 360°, tout en augmentant le sentiment de coprésence. En effet, chaque utilisateur distant ayant le point de vue de la caméra, tous se retrouvent localisés au même endroit sur le lieu d'intérêt. Le second verrou est de développer des interactions avec les données de la caméra 360°. Nous souhaitons particulièrement proposer aux utilisateurs distants de naviguer librement sur le lieu d'intérêt. Le problème est alors de trouver une représentation du lieu capable de générer plusieurs points de vue et qui peut être capturée avec une caméra 360° statique.Le manuscrit présente quatre contributions : un nouveau cadre théorique de la télé-immersion et trois versions de systèmes de télé-immersion basés sur une unique caméra 360° statique. La première exploite uniquement des images 360° sans informations 3D, la seconde intègre des informations 3D aux images 360° sous forme de cartes de profondeur, et la dernière profite d'une nouvelle représentation 3D 360°. Cette dernière version est évaluée avec une expérience utilisateur visant à montrer que le sentiment de présence qu'elle suscite est plus grand qu'avec la simple diffusion de la vidéo 360° capturée par la caméra
In contemporary professional environments, work is often dispersed across geographically distant locations. As organizing face-to-face meetings can be complex, professionals today rely on information and communication technologies to organize interactions. Tele-immersion is part of this technological trend, and aims to bring geographically separated individuals together as if they were present in the same place. This thesis proposes to develop an original tele-immersion system based on a 360° camera and extended reality headsets. The system is designed to enable users in a site of interest to bring distant users to them by simply placing a 360° camera. Thanks to extended reality, remote users are as if teleported to the place of interest. The system has been specifically developed to meet the need for new distance learning technologies, so that teachers can deliver immersive courses to students at home.The first challenge to achieve such a system is to manage the viewpoints of multiple users with a single 360° camera, while increasing the sense of copresence. In fact, as each remote user has the point of view of the camera, they are all located in the same place on the site of interest. The second challenge is to develop interactions with 360° camera data. In particular, we want to enable remote users to navigate freely around the place of interest. The problem is then to find a representation of the location capable of generating several points of view, and which can be captured with a static 360° camera.The thesis presents four contributions: a new theoretical framework for tele-immersion and three versions of tele-immersion systems based on a single static 360° camera. The first exploits only 360° images without 3D information, the second integrates 3D information into the 360° images in the form of depth maps, and the last takes advantage of a new 3D 360° representation. The latter version is evaluated with a user experience designed to illustrate that the sense of presence is greater than simply broadcasting the 360° video captured by the camera

La médecine nucléaire a connu ces dernières années d’importantes évolutions technologiques grâce à l’arrivée de caméras dites numériques, qui reposent sur l’utilisation de détecteurs semi-conducteurs. En tomographie d’émission monophotonique (TEMP), cette technologie a d’abord été introduite sur des caméras CzT dédiées à l’exploration cardiaque. Plus récemment, des caméras CzT 360° à champ de vue étendu ont été développées pour permettre des explorations tomographiques corps entier. De même, les caméras utilisées en tomographie par émission de positon (TEP) ont connu une transition vers des technologies numériques grâce à l’utilisation de nouveaux détecteurs SiPM. Ce travail a permis d’évaluer les performances de ces caméras TEMP et TEP numériques en médecine nucléaire dans le domaine de l’oncologie clinique et préclinique. Nous avons ainsi démontré les gains en sensibilité, en résolution en énergie et en contraste image obtenus à l’aide des caméras CzT 360° par rapport aux caméras de Anger conventionnelles. Il a également été démontré la faisabilité d’utiliser ce type de caméra, initialement développée pour l’homme, pour réaliser de l’imagerie chez le petit animal. Dans un second temps, nous avons évalué sur fantôme les caméras TEP SiPM pour une utilisation en préclinique. Nous avons ainsi mesuré des performances s’approchant de celles obtenues avec des caméras microTEP dédiées, en permettant l’imagerie simultanée de quatre animaux, et montré la possibilité de réaliser de la quantification avec une précision suffisante en oncologie préclinique
In recent years, nuclear medicine has undergone significant technological advances with the introduction of digital cameras based on the use of semiconductor detectors. In single-photon emission computed tomography (SPECT), this technology was first introduced on dedicated cardiac CzT cameras. More recently, 360° CzT cameras with extended field-of-view have been developed to enable whole-body tomographic explorations. Similarly, the cameras used in positron emission tomography (PET) have undergone a transition to digital technologies thanks to the use of new SiPM-based detectors. This work has allowed for the evaluation of the performance of these digital SPECT and PET cameras in nuclear medicine, within the field of clinical and preclinical oncology. We have thus demonstrated the improvements in sensitivity, energy resolution, and image contrast achieved through the use of 360° CzT cameras compared to conventional Anger cameras. Additionally, we demonstrated the feasibility of using this type of camera developed for humans, to perform preclinical imaging in small animals. Furthermore, we have evaluated SiPM-based PET cameras using phantoms for potential preclinical applications. We have thus measured performance approaching that obtained with dedicated microPET cameras, enabling simultaneous imaging of four animals, and demonstrated the ability to perform accurate quantification in preclinical oncology

In this research project, a system is proposed to aid the visually impaired by providing partial contextual information of the surroundings using 360° view camera combined with deep learning is proposed. The system uses a 360° view camera with a mobile device to capture surrounding scene information and provide contextual information to the user in the form of audio. The system could also be used for other applications such as logo detection which visually impaired users can use for shopping assistance.

The scene information from the spherical camera feed is classified by identifying objects that contain contextual information of the scene. That is achieved using convolutional neural networks (CNN) for classification by leveraging CNN transfer learning properties using the pre-trained VGG-19 network. There are two challenges related to this paper, a classification and a segmentation challenge. As an initial prototype, we have experimented with general classes such restaurants, coffee shops and street signs. We have achieved a 92.8% classification accuracy in this research project.

Thesis (M.S.) -- University of Maryland, College Park, 2006.
Thesis research directed by: Dept. of Fire Protection Engineering. Title from t.p. of PDF. Includes bibliographical references. Published by UMI Dissertation Services, Ann Arbor, Mich. Also available in paper.

Nowadays human motion analysis is one of the most active research topics in Computer Vision and it is receiving an increasing attention from both the industrial and scientific communities. The growing interest in human motion analysis is motivated by the increasing number of promising applications, ranging from surveillance, human–computer interaction, virtual reality to healthcare, sports, computer games and video conferencing, just to name a few. The aim of this thesis is to give an overview of the various tasks involved in visual motion analysis of the human body and to present the issues and possible solutions related to it. In this thesis, visual motion analysis is categorized into three major areas related to the interpretation of human motion: tracking of human motion using virtual pan-tilt-zoom (vPTZ) camera, recognition of human motions and human behaviors segmentation. In the field of human motion tracking, a virtual environment for PTZ cameras (vPTZ) is presented to overcame the mechanical limitations of PTZ cameras. The vPTZ is built on equirectangular images acquired by 360° cameras and it allows not only the development of pedestrian tracking algorithms but also the comparison of their performances. On the basis of this virtual environment, three novel pedestrian tracking algorithms for 360° cameras were developed, two of which adopt a tracking-by-detection approach while the last adopts a Bayesian approach. The action recognition problem is addressed by an algorithm that represents actions in terms of multinomial distributions of frequent sequential patterns of different length. Frequent sequential patterns are series of data descriptors that occur many times in the data. The proposed method learns a codebook of frequent sequential patterns by means of an apriori-like algorithm. An action is then represented with a Bag-of-Frequent-Sequential-Patterns approach. In the last part of this thesis a methodology to semi-automatically annotate behavioral data given a small set of manually annotated data is presented. The resulting methodology is not only effective in the semi-automated annotation task but can also be used in presence of abnormal behaviors, as demonstrated empirically by testing the system on data collected from children affected by neuro-developmental disorders.

Submitted by Ana Carla Almeida (ana.almeida@ucsal.br) on 2017-08-31T13:18:25Z No. of bitstreams: 1 DISSERTACAOKATHARINEMARQUES.pdf: 1568719 bytes, checksum: 9813ae289c1fad71cc11f91d0bc85bcc (MD5)
Approved for entry into archive by Rosemary Magalhães (rosemary.magalhaes@ucsal.br) on 2017-08-31T13:33:11Z (GMT) No. of bitstreams: 1 DISSERTACAOKATHARINEMARQUES.pdf: 1568719 bytes, checksum: 9813ae289c1fad71cc11f91d0bc85bcc (MD5)
Made available in DSpace on 2017-08-31T13:33:11Z (GMT). No. of bitstreams: 1 DISSERTACAOKATHARINEMARQUES.pdf: 1568719 bytes, checksum: 9813ae289c1fad71cc11f91d0bc85bcc (MD5) Previous issue date: 2017-03-29
É crescente a necessidade de rediscutir e repensar compreensões cada vez mais ampliadas sobre as relações intrafamiliares e os padrões de interação entre pais e filhos, como aspectos que interferem na criação dos filhos e na constituição de pessoas autônomas e colaborativas em seus diversos contextos. Nesse viés, o contexto familiar e as práticas educativas adotadas por pais e mães são aspectos centrais para o desenvolvimento da autonomia de seus filhos. Diante do exposto, temos como objeto de pesquisa a participação de pais e mães de adolescentes por meio de suas dinâmicas de interações vivenciadas em suas práticas educativas desempenhadas no exercício da parentalidade. Temos como problema de pesquisa: Como as práticas educativas de pais e mães da camada média urbana de Salvador contribuem no processo de desenvolvimento da autonomia de seus filhos? Para respondê-la, levantamos como objetivo geral: i) analisar a compreensão de pais e mães acerca de suas práticas educativas relacionadas ao desenvolvimento da autonomia de seu(s) filho(s) adolescente(s) e como específicos: ii) investigar as concepções de autonomia para pais e mães de filhos adolescentes; iii) compreender as práticas educativas de pais e mães relacionadas ao desenvolvimento da autonomia dos seus filhos adolescentes; e iv) discutir o papel parental no desenvolvimento da autonomia de filhos adolescentes na contemporaneidade. A metodologia utilizada foi de natureza qualitativa (MINAYO, 2010) e exploratória (YIN, 2010), sendo um estudo de casos múltiplos. Foram realizadas entrevistas abertas com três pais e três mães, casados entre si, do modelo de família nuclear. Aplicamos uma questão disparadora para permitir que os entrevistados respondessem de forma livre. Para a obtenção de dados complementares foi utilizado um roteiro de identificação com dados relacionados a cada participante. Posteriormente, as entrevistas foram analisadas com base na construção de categorias de análise de conteúdo (MINAYO, 2010). Foi feito uso de aportes teóricos sistêmicos (BOWEN, 1991; MINUCHIN, 1982, 2003, 2008; BRONFENBRENNER, 1996, 2011), com o intuito de fundamentar a análise e compreensão dos dados. Obtivemos como resultado relevante que o desenvolvimento da autonomia dos adolescentes é um processo contínuo, complexo e relacional e acontece na relação do indivíduo com a família e a sociedade mais ampla. Observamos que cada uma das famílias em estudo apresenta especificidades em relação às suas crenças e valores que medeiam o processo de educação dos filhos, como uma experiência única.
There is a growing need to rediscuit and rethink increasingly extended understandings about intrafamily relationships and patterns of parent-child interaction as aspects that interfere with parenting and the constitution of autonomous and collaborative people in their various contexts. In this bias, the family context and the educational practices adopted by fathers and mothers are central aspects for the development of the autonomy of their children. In view of the above, the present study has the objective of research the participation of parents and adolescents through their dynamics of interactions experienced in their educational practices performed in the exercise of parenting. We have as a research problem: How do the educational practices of urban middle-class parents contribute to the process of developing their children's autonomy? In order to answer this question, we have as a general objective: (i) to analyze the parents' understanding of their educational practices related to the development of the autonomy of their adolescent child (ren) and how specific: (ii) Conceptions of autonomy for fathers and mothers of adolescent children; Iii) understand the educational practices of parents related to the development of the autonomy of their adolescent children; And iv) to discuss the parental role in the development of the autonomy of adolescent children in contemporary times. The methodology used was qualitative (MINAYO, 2010) and exploratory (YIN, 2010), being a multiple case study. Open interviews were conducted with three parents and three mothers, married to each other, of the nuclear family model. We applied a triggering question to allow respondents to respond freely. To obtain additional data an identification roadmap was used with data related to each participant. Subsequently, the interviews were analyzed based on the construction of categories of content analysis (MINAYO, 2010). It was made use of systemic theoretical contributions (BOWEN, 1991; MINUCHIN, 1982, 2003, 2008; BRONFENBRENNER, 1996, 2011), in order to base the analysis and understanding of the data. We obtained as a relevant result that the development of adolescents' autonomy is a continuous process, complex and relational and happens in the relationship of the individual with the family and the wider society. We observed that each of the families studied presents specifics regarding their beliefs and values that mediate the process of raising their children as a unique experience.

Les caméras de surveillance font depuis quelques années partie de nos paysages urbains, voire même ruraux. Ces dispositifs n’ont pourtant pas réglé l’ensemble des problèmes qui ont amené à leurs déploiements dans de nombreux espaces publics. L’une des solutions proposées afin de régler ce défaut de la vidéosurveillance consiste alors en l’ajout de logiciels de détection automatique au système existant. Ces briques supplémentaires sont de différentes sortes : comptage de personnes, détection de types de véhicule, détection de dépôts sauvages, analyse de situations anormales, reconnaissance faciale etc. Les premiers usages de ces nouvelles formes de vidéosurveillance ont eu lieu en France, bien que celles-ci ne soient pas aussi nombreuses que dans d’autres pays. Cette thèse cherche à apporter des première réflexions concernant ces nouveaux outils de surveillance et le contexte qui les entoure en les abordant par trois aspects principaux. Le premier de ces aspects concerne les développements scientifiques qui participent de la construction de ces nouvelles solutions de surveillance. Le deuxième aspect se concentre sur le débat public autour de la vidéosurveillance automatisée et la forme particulière que prend ce débat. Le troisième et dernier aspect interroge les utilisations, expérimentales pour la plupart, qui ont lieu en France, les objectifs et les conditions de ces usages. Pour le premier aspect, ce travail tend à montrer l’importance particulière des infrastructures dans la réussite de ces outils d’intelligence artificielle. Que ce soit à l’étape du développement ou bien de l’usage, l’intelligence artificielle repose en grande partie sur tous outils, tous ces matériaux mobilisés dans son fonctionnement. Données, réseaux, capteurs sont autant de tamis par lesquels le travail du logiciel passe et qui imposent des limites techniques. Pour le second aspect, la recherche a exploré la question de la présentation de ces solutions techniques, de leur promotion ou de leur critique, en tant que problème, dans les débats publics ou bien dans les espaces économiques. Il en ressort que malgré une critique diffuse dans la société les arguments économiques dominent dans les rapports officiels. Pour le troisième aspect, alors que le sujet de la surveillance automatisée ne fait pas consensus, la thèse analyse l’ensemble des stratégies et ressources mobilisées afin de construire malgré tout un marché de la vidéosurveillance automatisée
CCTVs have been part of our urban and even rural landscapes for some years now. However, they have not solved all the problems that led to their deployment in many public spaces. One solution to this problem is to add automatic detection software to the existing system. These additional components come in a variety of forms: people counting, vehicle type detection, illegal dumping detection, analysis of abnormal situations, facial recognition, etc. The first uses of these new forms of video surveillance have taken place in France, although they are not as numerous as in other countries. This thesis seeks to provide some initial thoughts on these new surveillance tools and the context surrounding them, by approaching them from three main angles. The first concerns the scientific developments involved in the construction of these new monitoring solutions. The second focuses on the public debate surrounding automated video surveillance and the particular form this debate is taking. The third and final aspect examines the uses - most of them experimental - that are taking place in France, and the objectives and conditions of these uses. For the first aspect, this work tends to show the particular importance of infrastructures in the success of these artificial intelligence tools. Whether at the development stage or in use, artificial intelligence relies heavily on all the tools and materials mobilized for its operation. Data, networks and sensors are all sieves through which the software's work passes, and which impose technical limits. As for the second aspect, the research explored the question of how these technical solutions are presented, promoted or criticized as a problem, in public debates or in economic arenas. This revealed that, despite widespread social criticism, economic arguments dominate official reports. As for the third aspect, while there is no consensus on the subject of automated surveillance, the thesis analyzes all the strategies and resources mobilized to build an automated video surveillance market in spite of everything

In the post-war period, Italy has been plagued by different forms of organised crime (such as the Sicilian Mafia, the Neapolitan Camorra, the Calabrian 'Ndrangheta and the Pugliese Sacra Corona Unita) which have managed in their individual ways to infiltrate both the State (in the form of political parties and local administrations) and society (businesses, cultural and voluntary organisations). In Campania, until 1991, the relationship between the Camorra and the local political elite (based on the exchange of votes for state contracts and protection) was tacitly accepted by the population and could not be studied by political scientists due to the lack of reliable source material. In 1991, a law was introduced which gave generous remission of sentences to criminals who became state-witness. Many members of the Camorra revealed important aspects of criminal, economic and political activities in Campania. This new material permitted a reexamination of the Camorra. This thesis on the Camorra hopes to fill a gap in the study of the relationship between politics and criminal organisations which so far has concentrated on the Sicilian Mafia. Part One is a general introduction and presents the theoretical model and methods adopted. The documentation available allowed us to adopt an agency-structure approach derived from Giddens's structuration theory (1984). This was complemented by Easton's systems analysis (1965) to understand the wider, macro-environment. We elaborated an 'interaction model' to analyse the changing nature of the Camorra's activities: from a simple social-criminal practice in the 1930s to a dynamic and secret cartel enacting a political-criminal practice in the 1980s. To test this model we applied it to case-studies of criminals using original judicial documents. In Part Two we look at the possible motives of people who join the Camorra. We analysed the agent's internal and external structure in both decades and concluded that the macro-environment as an influencing factor had changed more than the individual-agent. Part Three examines the lives of camorristi in the 1950s and 1980s in order to determine how far their criminal practice has been transformed. Part Four investigates the wider picture of system-interaction between the Camorra's social subsystem and the political system. Focussing on the relationship between camorristi and the political elite in the 1950s and 1980s we highlight the radical changes that occurred. This thesis presents a theoretical discussion of how to study organised crime and social behaviour in general and at the same time a detailed empirical study, in particular of the political role of a criminal organisation in a concrete historical situation, that of Naples over the last forty years.

In den letzten Jahren kommen immer mehr nicht perspektivische Kamerasysteme beim maschinellen Sehen zur Anwendung, die vor allem ein deutlich erweitertes Blickfeld bieten. Das klassische perspektivische Abbildungsmodell lässt sich hier häufig nicht mehr erfolgreich anwenden. In dieser Arbeit wird ein generisches Abbildungsmodell vorgestellt, welches übliche Kamerasysteme akkurat modellieren kann. Solche Kamerasysteme schließen insbesondere klassische perspektivische Systeme, aber auch Fischaugen- und Spiegellinsen-Kamerasysteme ein. Die Nutzung eines einheitlichen Abbildungsmodells ermöglicht schließlich eine einfache Verwendung und Kalibrierung von heterogenen Stereokamerasystemen, also einer Kombination von unterschiedlichen Kameratypen, die vorteilhafte Eigenschaften gegenüber klassischen Stereosystemen bieten. Nicht zuletzt trägt die in dieser Arbeit vorgestellte einheitliche Modellierung und Kalibrierung von Mono- und Stereokamerasystemen dazu bei, Fehler durch falschen Umgang oder falsche Wahl von Methoden der Modellierung oder Kalibrierung zu vermeiden und den Kamerakalibrierprozess insgesamt zu vereinfachen. In dieser Arbeit wurden verschiedene Ansätze der Modellierung untersucht und evaluiert. Es wurde eine generische Modellierung vorgeschlagen, die die untersuchten spezifischen Abbildungsmodelle vollständig ersetzen kann. Für die Kalibrierung nicht linearer Abbildungsmodelle wurde eine einheitliche Methode zur Startwertbestimmung vorgeschlagen und evaluiert. Die Genauigkeit der Kalibrierung mittels einheitlicher Methoden wurde anhand diverser realer Kamerasysteme untersucht und bewertet. Es konnte gezeigt werden, dass die dabei auftretenden Fehler deutlich im Subpixelbereich liegen. Durch Erweiterung des klassischen Konzepts der Epipolargeometrie um die generische Abbildungsmodellierung konnten schließlich heterogene Stereokamerasysteme kalibriert und genaue Stereomodelle abgeleitet werden.
The application of perspective camera systems in photogrammetry and computer vision is state of the art. In recent years non-perspective and especially omnidirectional camera systems have increasingly been used in close-range photogrammetry tasks. In general, the perspective camera model, i.e. pinhole model, cannot be applied when using non-perspective camera systems. However, several camera models for different omnidirectional camera systems are proposed in literature. Using different types of cameras in a heterogeneous camera system may lead to an advantageous combination. The advantages of different camera systems, e.g. field of view and resolution, result in a new enhanced camera system. If these different kinds of cameras can be modeled, using a unified camera model, the total calibration process can be simplified. Sometimes it is not possible to give the specific camera model in advance. In these cases a generic approach is helpful too. Furthermore, a simple stereo reconstruction becomes possible when using a fisheye and a perspective camera for example. In this work camera models for perspective, wide-angle and omnidirectional camera systems were evaluated. A generic camera model were introduced that fully substitutes specific camera models. The crucial initialization of the model''s parameters is conducted using a new generic method that is independent of the particular camera system. The accuracy of this generic camera calibration approach is validated by the calibration of a dozen of real camera systems up to subpixel accuracy. Finally, it has been shown that a unified method of modeling, parameter approximation and calibration of interior and exterior orientation can be applied to a generic stereo system to derive precise 3D object data.

Bildsensoren mit integrierter Signalverarbeitung - sog. "Vision Chips" - ermöglichen die Ausführung ansonsten rechenintensiver Verarbeitungsschritte während oder unmittelbar nach der Bildaufnahme. Gegenüber konventionellen CMOS-Bildsensoren, die sich vor allem durch eine gute Bildqualität auszeichnen, werden die auszugebenden Daten bereits auf dem Chip auf relevante Informationen beschränkt und lediglich extrahierte Merkmale anstelle vollständiger Bilder ausgegeben. Vision Chips ermöglichen somit eine sehr hohe Bildwiederholrate bei gleichzeitig deutlich niedrigeren Anforderungen bzgl. der Übertragungsbandbreite und sind insbesondere für die Beobachtung sehr schneller Prozesse attraktiv. Obwohl das Konzept der gemeinsamen Betrachtung von Bildaufnahme und -verarbeitung bereits in den Anfangsjahren der Halbleiter-Bildsensoren aufgegriffen wurde, können die meisten beschriebenen Sensoren als Machbarkeitsnachweise für bestimmte Pixelzellen- bzw. Bildverarbeitungstechnologien betrachtet werden. So finden sich, bis auf den in der optischen Maus eingesetzten Sensor zur Bestimmung der Verschiebung relativ zum Untergrund, nur für sehr wenige Sensoren Hinweise auf einen kommerziellen Einsatz. Neben einer geringen optischen Auflösung und einer eingeschränkten Empfindlichkeit können der Verzicht auf integrierte Steuerwerke und die erhebliche Komplexität bzgl. der Programmierung als wesentliche Hindernisse für einen breiten Einsatz genannt werden. Im Rahmen dieser Arbeit werden wesentliche Beiträge zu der zum Einsatz von Vision Chips in realen Aufgabenstellungen erforderlichen Infrastruktur geliefert. So wird zur Ansteuerung der einzelnen Funktionseinheiten (Functional Unit, FU) zunächst das Konzept eines integrierten, Multi-ASIP (Application Specific Instruction-set Processor) basierten Steuerwerks erarbeitet, das durch die Bereitstellung mehrerer Kontrollflüsse die Ansteuerung paralleler FU ermöglicht. Die praktische Umsetzung des Konzepts in Hardware erfolgt als Bestandteil eines Vision-System-on-Chip (VSoC). Dieses verfügt gegenüber dem Stand der Technik über eine höhere Auflösung sowie eine größere Empfindlichkeit und bildet die Grundlage der weiteren Betrachtungen. Eine umfangreiche Simulationsumgebung ermöglicht Untersuchungen implementierter Algorithmen sowohl hinsichtlich zeitabhängiger Effekte als auch bzgl. der Auswirkung einzelner, in Bildaufnahme- und Verarbeitung gezielt eingebrachter Fehler und Nicht-Idealitäten. Die zum Betrieb des VSoC erforderliche Entwicklungs- und Kameraplattform ist sowohl für den Einsatz unter realen Bedingungen als auch zur Entwicklung von Bildverarbeitungsaufgaben geeignet und ermöglicht dabei die transparente Nutzung der Simulationsumgebung komplementär zur eigentlichen Hardware. Zur Erschließung der vom VSoC bereitgestellten Funktionalität für tatsächliche Aufgabenstellungen erfolgt die ganzheitliche Betrachtung einer Bildverarbeitungsaufgabe bestehend aus VSoC-basierter Vor- und konventioneller Nachverarbeitung in Form sog. "Vision Tasks". Zur Vereinfachung der Implementierung werden parametrierbare Skeletons bereitgestellt, in denen generelle Abläufe zur Bildaufnahme und -verarbeitung hinterlegt werden. Basierend auf den entwickelten Konzepten werden schließlich mehrere Anwendungsbeispiele umgesetzt.

Description d'une camera ccd et d'un nouvel instrument de spectroscopie multiouverture en temps reel: le systeme puma. Application a l'observation spectroscopique de l'amas lointain abell 370 et de l'amas proche coma. On montre que le contenu galactique des amas lointains est sensiblement different de celui des amas proches et que des processus efficaces font evoluer leurs galaxies en quelques milliards d'annees. Un catalogue est etabli pour chacun des amas et une analyse dynamique de l'amas coma est traitee

Der Name Klinik ist von dem altgriechischen Wort klíne abgeleitet, das einen Gegenstand – das Bett oder die Liege benennt. Jedoch wird mit Klinik zunächst eine Praxis bzw. eine Methode bezeichnet: die „Medizin am Krankenbett“. In dieser Arbeit wird untersucht, welche Bedeutungen, Wirkungen und Entwicklungen das Bett in und mit der Klinik entfaltete. Seine Form, Materialität und symbolischen Eigenschaften wirken auf die Klinik als Institution, als architektonisches und organisatorisches Gebilde sowie auf die Patient_innen und die Art und Weise ihrer Behandlung. Seit dem späten 18. Jh. ist das eiserne Bettgestell, belegt mit einer Person, konstitutiv für den ‚zivilisierten‘ Menschen und zieht in dieser Form in Europas Hospitäler ein. Das Krankenbett hat außerdem stützende, unterstützende und schützende Funktionen. In Gestalt der ‚Bettmaschine‘ verspricht es um 1800 humanitären Halt und Trost. Im 20. Jh. übernimmt das Dispositiv Sicherheit: Bettgitter sollen ‚wehrlose Kranke‘ vor dem Herausfallen aus dem Bett schützen. Die Abrechnung und Planung der Krankenhausbehandlung wird seit dem 19. Jh. mehr und mehr an die Einheit Bett geknüpft. Anfang des 20. Jh. wurde zudem die Idee des Fortschritts mit dem Krankenhausbett als zeitgemäßes Industrieprodukt nach Normvorgaben verbunden. Designprojekte der 1960er Jahre richteten sich besonders auf die Körper einer Nutzergruppe: der Pflegefachkräfte. Bettennot, Fachkräftemangel, der Drang nach Verkürzung der Verweildauer und daraus resultierende Rationalisierungen veränderten das Bett, und mit ihm die Abläufe im Krankenhaus und wie Patient_innen versorgt und behandelt werden. Sichtbar wird zudem, dass zwei Mobilisierungsprojekte im Widerstreit zueinander stehen: Auf der einen Seite die Bemühungen, das Krankenhausbett für die klinische Logistik beweglich zu machen und auf der anderen Seite, die im langen Prozess der Aktivierung des Verkehrsapparats Krankenhaus passivierten Patient_innen zu mobilisieren.
Although the word ‘clinic’ refers to a practice or a method of "bedside teaching", it derives from the ancient Greek word klíne, which describes an object – the bed or the couch. This research project examines the evolution of the hospital bed itself and the meanings and effects associated with its development. Its form, materiality and symbolic properties have affected the clinic as an institution and as an architectural and organizational structure, as well as the patients and the way they have been treated. In the 18th century, the hospital bed became a means to draw distinctions between wild or civilized, healthy or ill etc. Historically, the sickbed has taken on reclining, supportive and protective functions. Around 1800, in the form of a new “bed-machine”, it promised humanitarian support and consolation. In the 20th century, the apparatus of security took over, and bed rails were installed to protect ‘helpless patients’ from falling out of bed. Since the 19th century, the planning of hospital treatment, and the accounting behind it, has been linked more and more to the bed as a unit. At the beginning of the 20th century, the idea of progress transformed the hospital bed into a state-of-the-art industrial product with fixed standards. Steel tube furniture has established itself as peculiarly adaptable and as the site of medical treatment. Design projects in the 1960s focused on the bodies of one user group – the nursing staff. Hospital bed shortages, lack of nursing staff and pressure to shorten the length of stay resulted in a rationalization that changed the bed’s structure, and with it hospital procedures, and how patients are treated. Two significant and controversial mobilization projects linked to this development are efforts to make the hospital bed mobile for the sake of logistics, and, as a countermeasure, current efforts to re-mobilize patients who were made passive by the former process.

碩士
國立交通大學
多媒體工程研究所
107
Unlike using viewfinders as with traditional cameras, photographers of 360 videos do not have such a viewfinder during shooting. This makes video editing difficult and hinders the viewing experience of 360-degree videos. This work introduces handling methods for the 360-degree camera and proposes Director360, a 360 camera enhanced with two novel handling modes. The Pointer and FoV modes are designed to explicitly and implicitly utilize the 360 camera operator's attention in regard to omnidirectional scenery for shooting video footage. The Pointer mode lets users specify an object of interest by directly pointing the 360 cameras at the object; while our algorithm based on deep learning estimates the camera operator's 360 scene field-of-view in FoV mode. Herein, we detail the implementation and demonstrate the feasibility of the FoV mode based on deep-learning algorithms with an experiment. We present the Director360 Editor which incorporates the handling data to streamline the editing process. To understand how Director360 helps shooting and to edit 360 videos, a user study with four participants were recruited to create video storytelling in three target scenarios, and their feedback is reported.

碩士
國立中央大學
光電科學研究所碩士在職專班
100
This research, which is based on the patent M272096 of Republic of China, aimed to design the three million pixels camera of 60-degree angle of view by using three plastic lens and sensor OV3630. In the process of designing, the researcher did not eliminate the extremum of lens intentionally. Moreover, comparative illumination was raised by controlling off-axis number aperture and the curve of distortion through real ray high. Furthermore, on the quality of imaging, the researcher limited the real ray high of margin ray, which was aiming respectively to the X and Y direction, and that is accompanied by standard merit function. After optimizing, the full length of this lens is 7.45114 mm, comparative illumination is greater than 62%, and the absolute value of optical distortion was controlled no more than 2%; moreover, all the MTF within 113 lps/mm of spatial frequency is greater than 47%. The results of this device above-mentioned were all qualified to the target of imaging quality.

碩士
國立臺灣大學
資訊工程學研究所
106
In this thesis, we introduce a method about synthesizing stereo view for VR with 360 image sequence captured by a monocular 360-camera. We generate different 360 images for different eyes by referring to the ideas of Omnistereo. To generate a stereo view in VR, we combined the images，one for left, and the other for right eye. The images are captured by Gear 360, and then the three-dimensional points and poses were illustrated by using the structure from motion with Visual SfM. The three-dimensional points could work as the features of the images, and accordingly generate the features in new images by using multiple viewpoint circular projection. We use feature correspondences to synthesize novel views, while simultaneously maintaining stereoscopic properties and preserving image structures. Finally, we get the 360 images for left and right eyes. We use PsViewer to display our resulting images by setting different images for different eyes, and view in google cardboard.

碩士
國立交通大學
電子研究所
107
The 360 degree virtual view synthesis plays an important role in Virtual Reality and the depth map is the key information to reconstruct the 3D world. In this study, we use two spherical cameras to form a 360° stereo system, which can capture all the surrounding scene in two views. We then use these two spherical images to estimate the spherical depth map. We developed a depth estimation procedure on the spherical stereo images using an existing neural network, PSMNet. To train the network for spherical disparity estimation, we built a panorama stereo image dataset based on the SYNTHIA dataset, which has disparity ground truth. More importantly, we investigated the limits of spherical image depth estimation. Different from the disparity definition on the perspective view stereo, the spherical disparity is measured as the angle difference of the same object point on two views. Thus, the object aligned with the baseline has zero spherical disparity. Due to image plane pixel resolution, the maximum sensing distance for spherical disparity estimation was derived. Also, we studied the occlusion problem of a surface in spherical stereo, and derived the minimum reliable sensing distance. Both distance limits are functions of baseline. These properties help us in choosing an appropriate baseline length for constructing a spherical stereo. In our experiments, we performed depth estimation on both synthetic images and real scene images, and evaluated the performance on synthetic images with the ground truth depth. In the SYNTHIA test set, we can achieve an error rate of 2.18% using the KITTI benchmark D1 error criterion, which is lower than the original PSMNet tested on the KITTI dataset. At the end, we generated the synthetic views using Facebook 3D photo tools and our estimated depth maps. The good subjective quality of the synthesized images indicates that our estimated depth map is rather accurate.

A research report submitted to the faculty of Engineering and the built environment, University of the Witwatersrand, Johannesburg, in partial fulfilment of the requirements for the degree of Master of Science in Engineering. Johannesburg, August 2016
There has been much interest in using the Microsoft Xbox 360 Kinect cameras for visual servo control applications. It is a relatively cheap device with expected shortcomings. This work contributes to the practical considerations of using the Kinect for visual servo control applications. A comprehensive characterisation of the Kinect is synthesised from existing literature and results from a nonlinear calibration procedure. The Kinect reduces computational overhead on image processing stages, such as pose estimation or depth estimation. It is limited by its 0.8m to 3.5m practical depth range and quadratic depth resolution of 1.8mm to 35mm, respectively. Since the Kinect uses an infra-red (IR) projector, a class one laser, it should not be used outdoors, due to IR saturation, and objects belonging to classes of non- IR-friendly surfaces should be avoided, due to IR refraction, absorption, or specular reflection. Problems of task stability due to invalid depth measurements in Kinect depth maps and practical depth range limitations can be reduced by using depth map preprocessing and activating classical visual servoing techniques when Kinect-based approaches are near task failure.
MT2017

abstract: Generating real-world content for VR is challenging in terms of capturing and processing at high resolution and high frame-rates. The content needs to represent a truly immersive experience, where the user can look around in 360-degree view and perceive the depth of the scene. The existing solutions only capture and offload the compute load to the server. But offloading large amounts of raw camera feeds takes longer latencies and poses difficulties for real-time applications. By capturing and computing on the edge, we can closely integrate the systems and optimize for low latency. However, moving the traditional stitching algorithms to battery constrained device needs at least three orders of magnitude reduction in power. We believe that close integration of capture and compute stages will lead to reduced overall system power. We approach the problem by building a hardware prototype and characterize the end-to-end system bottlenecks of power and performance. The prototype has 6 IMX274 cameras and uses Nvidia Jetson TX2 development board for capture and computation. We found that capturing is bottlenecked by sensor power and data-rates across interfaces, whereas compute is limited by the total number of computations per frame. Our characterization shows that redundant capture and redundant computations lead to high power, huge memory footprint, and high latency. The existing systems lack hardware-software co-design aspects, leading to excessive data transfers across the interfaces and expensive computations within the individual subsystems. Finally, we propose mechanisms to optimize the system for low power and low latency. We emphasize the importance of co-design of different subsystems to reduce and reuse the data. For example, reusing the motion vectors of the ISP stage reduces the memory footprint of the stereo correspondence stage. Our estimates show that pipelining and parallelization on custom FPGA can achieve real time stitching.
Dissertation/Thesis
Prototype
Masters Thesis Electrical Engineering 2018

Research Doctorate - Doctor of Philosophy (PhD)
In today’s world, cardiovascular diseases such as heart attacks and stoke are the leading type of chronic diseases in terms of premature death. Unhealthy diet is among habits that increase the metabolic risk in an individual (e.g., obesity, increased blood glucose, and raised blood pressure) that may lead to cardiovascular diseases. Therefore, being able to accurately monitor dietary intake activities of individuals could play an important part in promoting a healthier diet and preventing cardiovascular diseases. Wearable motion tracking sensors, inertial sensors in particular, are now commonly used in commercial grade wearables such as smartwatches and fitness bands. These sensors can be used to provide the biometric data required to monitor dietary intake. Since an intake activity mainly consists of a series of intake-associated gestures, automatic intake gesture detection forms the basis required to achieve automatic monitoring of dietary intake. The overarching goal of this research is to design machine learning models to enhance automatic intake gesture detection process and improve the performance of detection using data recorded from wrist-worn inertial motion tracking sensors. To lay a robust groundwork to achieve the goal, we conducted a systematic literature review to synthesise research on assessing dietary intake using upper limb motion tracking sensors. Our literature review revealed that wrist-worn inertial sensors are the most used sensors in intake gesture detection. It also became evident that deep learning is gaining more attention and showing more promise than classical machine learning algorithms in this field. Therefore, we chose to utilise deep learning models using data recorded from wrist-worn inertial sensors. To provide adequate training data for our deep learning models on the one hand and to facilitate research in this fields for all researchers around the gobble on the other hand, we conducted a series of data collection experiments in two phases to collect inertial and video intake gesture data. As a result, we made the OREBA (Objectively Recognizing Eating Behaviour and Associated Intake) multimodal datasets publicly available through publishing a paper. The OREBA datasets consists of data recorded from discrete dish data collection scenario in phase one (i.e., OREBA-DIS) and shared dish data collection scenario in phase two (i.e., OREBA-SHA). In our next study, we used the OREBA datasets to propose a new deep learning model that improved the performance of intake gesture detection compared to the existing state-of-the-art models. We also clarified the effects of data preprocessing steps and inertial sensor combinations on intake gesture models. Our results showed that applying a consecutive combination of mirroring, removing gravity effect, and standardization data preprocessing steps improved the performance of detection, while smoothing had detrimental impact on the performance. In our last study, we explored the possibility of score-level and decision-level fusion of inertial and video data using the OREBA multimodal datasets. We benchmarked score-level and decision-level fusion approaches against no fusion (i.e., using individual inertial model or individual video model) approaches. Our results showed that the score-level fusion approach of max score model outperforms all other fusion approaches considered. However, it may not always outperform the no fusion approaches. We introduced an assessment that can be performed to determine the potential of score-level and decision-level fusion (i.e., fusion of outputs of the individual inertial and video models). Our assessment showed that fusion of the outputs of individual models is more promising on the OREBA-DIS dataset compared to OREBA-SHA dataset.