Au cours des dernières années les Mondes Virtuels 3D (MV3D) sont devenus une réalité. Initialement considérés comme un nouveau moyen de communication sociale, initiés par le développement logiciel et matériel, les MV3Ds révèlent diverses fonctionnalités, des expériences et des connaissances. Étant la représentation de l'utilisateur, l'avatar est l’une des ressources les plus significatives et les plus complexes d'un monde virtuel. Une courte analyse d'un contenu de MV nous informe que le stockage/transmission est la partie la plus significative des ressources du MV. Dans l’ensemble des ressources, les avatars sont les structures les plus complexes, représentés par différents composants : géométrie, images, animations, etc. Donc, en abordant les problématiques de compression, d'adaptation et d'interopérabilité liées aux avatars, nous traitons implicitement presque tout type de ressources tifs qu’on pourrait retrouver dans les MVs. L'objectif global du développement d’outils et des méthodes, pour un déploiement élargi de MV, est traduit par les trois points spécifiques suivants: Proposer un framework de compression pour permettre le transfert efficace et compact d'avatars et de ressources graphiques 3D généraux. Spécifiquement pour des avatars, le framework devrait être indépendante avec un formalisme de représentation. Proposer une solution optimisée permettant l’accessibilité aux avatars sur des terminaux à faibles ressources tels que les téléphones portables. Définir un modèle de métadonnées permettant l'interopérabilité d'avatars entre différents MVs. Les trois objectifs sont traités dans ce manuscrit et pour chacun nous proposons des contributions originales. / In the last few years 3D Virtual Worlds (3DVWs) became a reality. Initially considered as a new mean for social communication, triggered by the development of software and hardware technology, 3DVWs are exposing now different functionalities, experiences and acquaintances. Therefore, they achieved their popularity very fast, indicated by the number and the progression of active users. Being the representation of the user, the avatar is one of the most significant and most complex assets of a Virtual World. A short analysis of a VW content allows one to observe that from the point of view of the storage/transmission the most significant amount is represented by the VW assets. Within the set of assets, the avatars are the most complex structures, consisting of different components: geometry, images, animations, structures, etc. The overall objective of developing tools and methods for a large deployment of VW are translated into three specific ones: To propose a compression framework to enable efficient, compact transfer of avatars, and general 3D graphics assets. Specifically for avatars, the framework should be independent from the representation formalism. To propose an optimized solution making the avatars accessible on weak terminals such as mobile phones. To define a metadata model allowing avatars interoperability between different VWs. The three objectives are addressed in this thesis and for each we propose original contributions.

In the last few years 3D Virtual Worlds (3DVWs) became a reality. Initially considered as a new mean for social communication, triggered by the development of software and hardware technology, 3DVWs are exposing now different functionalities, experiences and acquaintances. Therefore, they achieved their popularity very fast, indicated by the number and the progression of active users. Being the representation of the user, the avatar is one of the most significant and most complex assets of a Virtual World. A short analysis of a VW content allows one to observe that from the point of view of the storage/transmission the most significant amount is represented by the VW assets. Within the set of assets, the avatars are the most complex structures, consisting of different components: geometry, images, animations, structures, etc. The overall objective of developing tools and methods for a large deployment of VW are translated into three specific ones: To propose a compression framework to enable efficient, compact transfer of avatars, and general 3D graphics assets. Specifically for avatars, the framework should be independent from the representation formalism. To propose an optimized solution making the avatars accessible on weak terminals such as mobile phones. To define a metadata model allowing avatars interoperability between different VWs. The three objectives are addressed in this thesis and for each we propose original contributions.

Nowadays, new computers generation provides a high performance that enables to build computationally expensive computer vision applications applied to mobile robotics. Building a map of the environment is a common task of a robot and is an essential part to allow the robots to move through these environments. Traditionally, mobile robots used a combination of several sensors from different technologies. Lasers, sonars and contact sensors have been typically used in any mobile robotic architecture, however color cameras are an important sensor due to we want the robots to use the same information that humans to sense and move through the different environments. Color cameras are cheap and flexible but a lot of work need to be done to give robots enough visual understanding of the scenes. Computer vision algorithms are computational complex problems but nowadays robots have access to different and powerful architectures that can be used for mobile robotics purposes. The advent of low-cost RGB-D sensors like Microsoft Kinect which provide 3D colored point clouds at high frame rates made the computer vision even more relevant in the mobile robotics field. The combination of visual and 3D data allows the systems to use both computer vision and 3D processing and therefore to be aware of more details of the surrounding environment. The research described in this thesis was motivated by the need of scene mapping. Being aware of the surrounding environment is a key feature in many mobile robotics applications from simple robotic navigation to complex surveillance applications. In addition, the acquisition of a 3D model of the scenes is useful in many areas as video games scene modeling where well-known places are reconstructed and added to game systems or advertising where once you get the 3D model of one room the system can add furniture pieces using augmented reality techniques. In this thesis we perform an experimental study of the state-of-the-art registration methods to find which one fits better to our scene mapping purposes. Different methods are tested and analyzed on different scene distributions of visual and geometry appearance. In addition, this thesis proposes two methods for 3d data compression and representation of 3D maps. Our 3D representation proposal is based on the use of Growing Neural Gas (GNG) method. This Self-Organizing Maps (SOMs) has been successfully used for clustering, pattern recognition and topology representation of various kind of data. Until now, Self-Organizing Maps have been primarily computed offline and their application in 3D data has mainly focused on free noise models without considering time constraints. Self-organising neural models have the ability to provide a good representation of the input space. In particular, the Growing Neural Gas (GNG) is a suitable model because of its flexibility, rapid adaptation and excellent quality of representation. However, this type of learning is time consuming, specially for high-dimensional input data. Since real applications often work under time constraints, it is necessary to adapt the learning process in order to complete it in a predefined time. This thesis proposes a hardware implementation leveraging the computing power of modern GPUs which takes advantage of a new paradigm coined as General-Purpose Computing on Graphics Processing Units (GPGPU). Our proposed geometrical 3D compression method seeks to reduce the 3D information using plane detection as basic structure to compress the data. This is due to our target environments are man-made and therefore there are a lot of points that belong to a plane surface. Our proposed method is able to get good compression results in those man-made scenarios. The detected and compressed planes can be also used in other applications as surface reconstruction or plane-based registration algorithms. Finally, we have also demonstrated the goodness of the GPU technologies getting a high performance implementation of a CAD/CAM common technique called Virtual Digitizing.

Dans cette thèse, nous identifions et exploitons des similarités partielles dans des objets 3D pour répondre à des besoins courants du domaine de la Conception Assistée par Ordinateur (CAO). De nouvelles méthodes sont introduites, d'une part pour détecter les similarités partielles, d'autre part pour utiliser ces similarités dans des applications spécifiques telles que l'édition de forme, la compression et l'indexation d'objets 3D. Grâce au développement des applications de la modélisation géométrique, ces modèles sont de plus en plus nombreux et sont disponibles à travers plusieurs modalités. Pour augmenter la productivité dans la création de tels objets virtuels, la réutilisation et l'adaptation des modèles existants est un choix prioritaire. Cela exige donc des méthodes facilitant le stockage, la recherche et l'exploitation de ces modèles. Heureusement, les similarités dans des objets 3D est un phénomène fréquent. De nombreux objets sont composés de parties similaires à une rotation, à une translation ou à une symétrie près. De ce fait, la détection des similarités partielles dans ces modèles est capable de répondre aux problématiques courantes : la taille du stockage est réduite en conservant seulement une partie au lieu de toutes les parties répétées d'un modèle; l'indexation des modèles 3D requiert a priori l'orientation canonique des modèles. Or, la symétrie dans un objet 3D est toujours une référence d'orientation cohérente avec la perception humaine. Nous utilisons donc la symétrie partielle pour aligner ces modèles et ainsi renforcer la robustesse des méthodes d'indexation. Dans un premier temps, nous introduisons une approche similaire à la Transformée de Hough pour détecter des similarités partielles dans des modèles BRep-NURBS. Cette approche identifie non seulement les parties similaires mais aussi les transformations qui les lient. À travers la classification des isométries dans l'analyse des transformations, notre approche peut distinguer la nature de transformation liant des parties similaires d'un modèle, c'est-à-dire, les parties similaires à une rotation, à une translation ou à une symétrie près. Dans le deuxième temps, nous proposons deux applications héritées directement des résultats obtenus par la détection. Tout d'abord, pour la compression, un modèle se transforme en un graphe de similarités d'où les faces principales à conserver sont sélectionnées dans la structure compressée. Ensuite, pour l'orientation, le plan de la symétrie dominante et la projection orthographique d'un modèle autour de ce plan permettent de définir un repère canonique pour aligner ce modèle. / In this thesis, we identify and exploit the partial similarities within 3D objects to answer the current needs of the Computer Aided Design field (CAD). Novel methods are introduced, on the one hand to detect the partial similarities, on the other hand to use these similarities for specific applications such as shape editing, compression and indexation of 3D objects. Because of the development of geometric modeling applications, 3D models are getting more numerous and available through many channels. To increase the productivity in creating such 3D virtual objects, the reuse and the adaptation of existing models becomes a prior choice. Thus, it requires methods easing the storage, the searching and the exploitation of these models. Fortunately, similarities within the 3D objects is a popular phenomenon. Many objects are composed of similar patches up to an approximated rotation, translation or symmetry. Hence, detecting the partial similarities within NURBSBRep models is able to solve the current issues : the storage size is reduced by coding a single patch instead of repeated patches of a model ; 3D model indexation requires a canonical orientation of these models. Furthermore, the symmetry within a 3D object is a good orientation reference, coherent with the human perception. Accordingly, we use the partial symmetries to align 3D models and so reinforce the robustness of indexation methods. In a first phase, we introduce an orginal approach similar to the Hough Transform to detect partial similarities within NURBS-BRep models. This approach identifies not only similar patches but also identifies the corresponding transformations that connect them. Additionally, through the classification of isometries in transformations analysis, our approach can distinguish the nature of transformations of similar patches of a model, that is, the patches similar up to an approximated rotation, translation or symmetry. This classification is advantageous for further applications : the similar patches of other transformation natures are considered in compressing ; the symmetric patches are used to normalize 3D models aim at a robust indexation. In the second phase, we propose two applications inherited directly from the obtained results of the detection. Firstly, for the compression, a model is transformed into a similarity graph where the principal faces to be coded are selected to form the compressed structure. Secondly, for the orientation, the plane of the dominant symmetry and the orthographic projection of a model around this plane generate a canonical frame to align this model.