In this thesis, we study the shape analysis problem and propose new methods for shape description, similarity and recognition. Firstly, we introduce a new shape descriptor in a two-step method. In the first step, the 2-D shape information is mapped into a set of 1-D functions. The mapping is based on the beams, which are originated from a boundary point, connecting that point with the rest of the points on the boundary. At each point, the angle between a pair of beams is taken as a random variable to define the statistics of the topological structure of the boundary. The third order statistics of all the beam angles is used to construct 1-D Beam Angle Statistics (BAS) functions. In the second step, we apply a set of feature extraction methods on BAS functions in order to describe it in a more compact form. BAS functions eliminate the context-dependency of the representation to the data set. BAS function is invariant to translation, rotation and scale. It is insensitive to distortions. No predefined resolution or threshold is required to define the BAS functions. Secondly, we adopt three different similarity distance methods defined on the BAS feature space, namely, Optimal Correspondence of String Subsequences, Dynamic Warping and Cyclic Sequence Matching algorithms. Main goal in these algorithms is to minimize the distance between two BAS features by allowing deformations. Thirdly, we propose a new Hidden Markov Model (HMM)topology for boundary based shape recognition. The proposed topology called Circular HMM is both ergodic and temporal. Therefore, the states can be revisited in finite time intervals while keeping the sequential information in the string, which represents the shape. It is insensitive to size changes. Since it has no starting and terminating state, it is insensitive to the starting point of the shape boundary. Experiments are done on the dataset of MPEG 7 Core Experiments Shape-1. It is observed that BAS descriptor outperforms all the methods in the literature. The Circular HMM gives higher recognition rates than the classical topologies in shape analysis applications.

In this thesis, we introduce a new shape descriptor and a graph based matching algorithm to detect a template shape in an image that contains a single object. The shape descriptor, Generalized Beam Angle Statistics, GBAS is obtained with the generalization of the boundary based shape descriptor, Beam Angle Statistics, BAS cite{BAS}. GBAS improves BAS so that it can compute the feature vector of a boundary point without the requirement of the parametric boundary representation. This way, it can be used in matching an individual edge pixel with a boundary point of template shape, even if it is not possible to extract the shape boundary in the image with the available techniques. Given a template shape, the matching algorithm solves the correspondence problem between the sampled boundary points of the template and the edges of the query image, using the GBAS feature vectors and the spatial information of edges. The match graph represents the correspondence problem and the optimum path on this graph gives the solution of it. Optimum path is found using a polynomial time algorithm that is based on the dynamic programming approach. In the experiments, we show that the proposed shape descriptor is very powerful and the matching algorithm is capable of detecting a template shape in edge detected images under a variety of transformations and noise.

<p>The recent development of Virtual Reality technology, researchers are looking more into changing the way Virtual Reality is used in our daily lives in order to increase our productivity. One such application is the mapping of 3D spatial graphics in Computer Aided Design engineering where practitioners have been historically working on 3D models in a two dimensional environment. Researchers in Computer Graphics have proposed Virtual Reality as a more effective medium for CAD packages. This thesis carries out a user study to test whether or not 3D VR environments are more effective in relaying information to the users as compared to two dimensional displays such as computer screens by conducting a study to determine how users navigate and interact with complex CAD objects in the two different environments. The two environments make use of stereoscopic vision and monoscopic vision in order to compare the efficiency with which volunteers are able to notice subtle differences in objects. The motivation for this study stems from the fact that CAD in VR is largely an underdeveloped topic and the result of such a study could form a baseline and advocate for further research and development in this domain. The research question being addressed is “Does CAD in a three-dimensional Virtual Reality Environment(stereoscopic) allow for better understanding of shapes of complex assemblies as compared to CAD on two-dimensional (monoscopic) computer screens?” The findings of this study suggest that rather than just the display technique the kind of movements which objects undergo also contributes to the way users perceive the objects in 3D vs 2D spaces and uncover a set of directions which would be recommended for similar studies in the future.</p><div><br></div>

[pt] Há uma crescente demanda pela criação de modelos computacionais representativos de objetos reais para projetos de engenharia. Uma alternativa barata e eficaz consiste na utilização de técnicas de Visão Computacional baseada em câmeras e projetores disponíveis no mercado de computadores pessoais. Este trabalho avalia um sistema óptico estéreo ativo para capturar formas geométricas de objetos utilizando um par de câmeras e um projetor digital. O sistema se baseia em idéias de trabalhos anteriores, com duas contribuições nesta dissertação. A primeira é uma técnica mais robusta de detecção de pontos notáveis em padrões de calibração das câmeras. A segunda contribuição consiste num novo método de ajuste de cilindros que visa aplicar o sistema estudado na inspeção de instalações de dutos industriais. As conclusões apresentadas procuram avaliar a robustez e precisão do sistema proposto como um instrumento de medidas em Engenharia. / [en] There has been a growing demand for creation of computer models based on real models for engineering projects. A cheap and effective alternative consists in using Computer Vision techniques based on cameras and projectors available at the personal computer market. This work evaluates a stereo optic system for capturing geometric shapes from objects using a pair of cameras and a single digital projector. The system is based on former works and a pair of contributions is obtained at this dissertation. The first contribution is a more robust technique for finding corners and points at cameras calibration patterns. And the second one consists on a new method for cylinder fit for inspecting industrial piping facilities with the studied system. The final conclusions evaluate the robustness and precision from the proposed system as a measurement tool for Engineering.

Cette thèse porte sur la reconnaissance et l’appariement de formes 3D pour des systèmes intelligents de vision par ordinateur. Elle décrit deux contributions principales à ce domaine. La première contribution est une implémentation d'un nouveau descripteur de formes construit à la base de la géométrie spectrale de l'opérateur de Laplace-Beltrami ; nous proposons une signature de point globale avancée (AGPS). Ce descripteur exploite la structure intrinsèque de l'objet et organise ses informations de manière efficace. De plus, AGPS est extrêmement compact puisque seulement quelques paires propres étaient nécessaires pour obtenir une description de forme précise. La seconde contribution est une amélioration de la signature du noyau d'onde ; nous proposons une signature du noyau d'onde optimisée (OWKS). La perfectionnement est avec un algorithme heuristique d'optimisation par essaim de particules modifié pour mieux rapprocher une requête aux autres formes appartenant à la même classe dans la base de données. L'approche proposée améliore de manière significative la capacité discriminante de la signature. Pour évaluer la performance de l'approche proposée pour la récupération de forme 3D non rigide, nous comparons le descripteur global d'une requête aux descripteurs globaux du reste des formes de l'ensemble de données en utilisant une mesure de dissimilarité et trouvons la forme la plus proche. Les résultats expérimentaux sur différentes bases de données de formes 3D standards démontrent l'efficacité des approches d'appariement et de récupération proposées par rapport aux autres méthodes de l'état de l'art. / This thesis concerns recognition and matching of 3D shapes for intelligent computer vision systems. It describes two main contributions to this domain. The first contribution is an implementation of a new shape descriptor built on the basis of the spectral geometry of the Laplace-Beltrami operator; we propose an Advanced Global Point Signature (AGPS). This descriptor exploits the intrinsic structure of the object and organizes its information in an efficient way. In addition, AGPS is extremely compact since only a few eigenpairs were necessary to obtain an accurate shape description. The second contribution is an improvement of the wave kernel signature; we propose an optimized wave kernel signature (OWKS). The refinement is with a modified particle swarm optimization heuristic algorithm to better match a query to other shapes belonging to the same class in the database. The proposed approach significantly improves the discriminant capacity of the signature. To assess the performance of the proposed approach for nonrigid 3D shape retrieval, we compare the global descriptor of a query to the global descriptors of the rest of shapes in the dataset using a dissimilarity measure and find the closest shape. Experimental results on different standard 3D shape benchmarks demonstrate the effectiveness of the proposed matching and retrieval approaches in comparison with other state-of-the-art methods.

Orientador: Elza Guimarães / Resumo: A florivoria pode ter um impacto significativo na reprodução das espécies vegetais devido à modificação da forma, odor ou padrão de coloração das flores, o que pode comprometer a comunicação e a interação entre flores e seus polinizadores. No caso dos beija-flores, danos nas flores e a descaracterização da forma floral podem representar importantes interferências na comunicação planta-polinizador, uma vez que os beija-flores são animais que se guiam essencialmente pela visão e pela memória. Neste trabalho, apresentamos dois capítulos em que exploramos o impacto da florivoria sobre a polinização de espécies ornitófilas. No primeiro, trazemos um estudo em que investigamos se danos semelhantes aos causados por florívoros em Pyrostegia venusta (Bignoniaceae), podem interferir na visita dos polinizadores. Já no segundo capítulo, trazemos um estudo mais amplo, envolvendo seis espécies ornitófilas, e investigamos se os beija-flores deixam de visitar flores danificadas, e se há uma preferência em relação às flores íntegras, visando explorar mais a fundo a importância da integridade da forma floral para a polinização por beija-flores. Em ambos os estudos, verificamos baixos indices de florivoria em todas as espécies estudadas. Além disso, obtivemos o mesmo resultado no que diz respeito à influência dos danos às flores (florivoria) sobre a polinização ornitófila, concluindo que a integridade floral não é essencial para a manutenção da interação planta-polinizador em espécies ornitófila... (Resumo completo, clicar acesso eletrônico abaixo) / Mestre

Il existe une multitude de traitements d'images dans la littérature, chacun étant adapté à un ensemble plus ou moins grand de cadres d'application. Les traitements d'images sont fondamentalement trop différents les uns par rapport aux autres pour être mis en commun de façon naturelle. De plus, ces derniers sont trop rigides pour pouvoir s'adapter d'eux mêmes lorsqu'un problème non prévu à l'avance par le concepteur apparaît. Or la vision est un phénomène autoadaptatif, qui sait traiter en temps réel des situations singulières, en y proposant des traitements particuliers et adaptés. Elle est aussi un traitement complexe des informations, tant ces dernières ne peuvent être réduites à des représentations réductionnistes et simplifiantes sans être mutilées.Dans cette thèse, un système de vision est entrepris comme un tout où chaque partie est adaptée à l'autre, mais aussi où chaque partie ne peut s'envisager sans l'autre dans les tensions les plus extrêmes générées par la complexité et l'intrication des informations. Puisque chaque parcelle d'information joue un rôle local dans la vision, tout en étant dirigée par un objectif global peu assimilable à son niveau, nous envisageons la vision comme un système où chaque agent délibère selon une interférence produite par le potentiel décisionnel de chacun de ses voisins. Cette délibération est entreprise comme le résultat produit par l'interférence d'une superposition de solutions. De cette manière, il émerge du système à base d'agents une décision commune qui dirige les actions locales faites par chaque agent ou chaque partie du système. En commençant par décrire les principales méthodes de segmentation ainsi que les descripteurs de formes, puis en introduisant les systèmes multi-agents dans le domaine de l'image, nous discutons d'une telle approche où la vision est envisagée comme un système multi-agent apte à gérer la complexité inhérente de l'information visuelle tant en représentation qu'en dynamisme systémique. Nous encrons dans ces perspectives deux modèles multi-agents. Le premier modèle traite de la segmentation adaptative d'images sans calibration manuelle par des seuils. Le deuxième modèle traite de la représentation de formes quelconques à travers la recherche de coefficients d'ondelettes pertinents. Ces deux modèles remplissent des critères classiques liés au traitement d'images, et à la reconnaissance de formes, tout en étant des cas d'études à développer pour la recherche d'un système de vision auto-adaptatif tel que nous le décrivons. / Although several image processing approaches exist, each of them was introduced in order to be used in a specific set of applications. In fact, image processing algorithms are fundamentally too different in order to be merged in a natural way. Moreover, due to their rigidity, they are unable to adapt themselves when a non-previously programmed problem appears as it could be the case in our framework. Indeed vision is an auto-adaptive phenomenon which can deal with singular situations by providing particular and adapted treatments. It is also a complex information processing. Therefore, vision should not be reduced to reductionist and simplifying representation. According to this thesis, a vision system could be developed as a whole in which each part adapts itself with others. Its parts cannot be considered separately due to the extreme tensions generated by the complexity and the intricacy of information. Each of them contributes locally to the vision and it is directed by a global objective incomprehensible at its level. We consider vision as a system whose agents deliberate according to an interference produced by the decision potential of each agent. This deliberation is undertaken as the result produced by interferences of a solution superposition. Then, it emerges from the agent-based system a common decision which directs local actions of each agent or of each part of the system. After describing the main shape descriptors and segmentation algorithms and after introducing multi-agent systems on the image processing domain, we discuss on approaches for which vision is considered as a multi-agent system able to manage the inherent complexity of visual information. Then, we give two multi-agent models. The first one deals with an adaptive segmentation which doesn't need manual calibration through thresholds. The second one deals with shape representations through the search of pertinent wavelet coefficients. These two models respect classical image processing criteria. They also are case studies that should be developed in the search of an auto-adaptive vision system.

Magister Scientiae - MSc / Research has shown that five fundamental parameters are required to recognize any sign language gesture: hand shape, hand motion, hand location, hand orientation and facial expressions. The South African Sign Language (SASL) research group at the University of the Western Cape (UWC) has created several systems to recognize sign language gestures using single parameters. These systems are, however, limited to a vocabulary size of 20 – 23 signs, beyond which the recognition accuracy is expected to decrease. The first aim of this research is to investigate the use of two parameters – hand motion and hand shape – to recognise a larger vocabulary of SASL gestures at a high accuracy. Also, the majority of related work in the field of sign language gesture recognition using these two parameters makes use of Hidden Markov Models (HMMs) to classify gestures. Hidden Markov Support Vector Machines (HM-SVMs) are a relatively new technique that make use of Support Vector Machines (SVMs) to simulate the functions of HMMs. Research indicates that HM-SVMs may perform better than HMMs in some applications. To our knowledge, they have not been applied to the field of sign language gesture recognition. This research compares the use of these two techniques in the context of SASL gesture recognition. The results indicate that, using two parameters results in a 15% increase in accuracy over the use of a single parameter. Also, it is shown that HM-SVMs are a more accurate technique than HMMs, generally performing better or at least as good as HMMs.

The work done in this master's thesis, presents a new system for the recognition of human actions from a video sequence. The system uses, as input, a video sequence taken by a static camera. A binary segmentation method of the the video sequence is first achieved, by a learning algorithm, in order to detect and extract the different people from the background. To recognize an action, the system then exploits a set of prototypes generated from an MDS-based dimensionality reduction technique, from two different points of view in the video sequence. This dimensionality reduction technique, according to two different viewpoints, allows us to model each human action of the training base with a set of prototypes (supposed to be similar for each class) represented in a low dimensional non-linear space. The prototypes, extracted according to the two viewpoints, are fed to a $K$-NN classifier which allows us to identify the human action that takes place in the video sequence. The experiments of our model conducted on the Weizmann dataset of human actions provide interesting results compared to the other state-of-the art (and often more complicated) methods. These experiments show first the sensitivity of our model for each viewpoint and its effectiveness to recognize the different actions, with a variable but satisfactory recognition rate and also the results obtained by the fusion of these two points of view, which allows us to achieve a high performance recognition rate. / Le travail mené dans le cadre de ce projet de maîtrise vise à présenter un nouveau système de reconnaissance d’actions humaines à partir d'une séquence d'images vidéo. Le système utilise en entrée une séquence vidéo prise par une caméra statique. Une méthode de segmentation binaire est d'abord effectuée, grâce à un algorithme d’apprentissage, afin de détecter les différentes personnes de l'arrière-plan. Afin de reconnaitre une action, le système exploite ensuite un ensemble de prototypes générés, par une technique de réduction de dimensionnalité MDS, à partir de deux points de vue différents dans la séquence d'images. Cette étape de réduction de dimensionnalité, selon deux points de vue différents, permet de modéliser chaque action de la base d'apprentissage par un ensemble de prototypes (censé être relativement similaire pour chaque classe) représentés dans un espace de faible dimension non linéaire. Les prototypes extraits selon les deux points de vue sont amenés à un classifieur K-ppv qui permet de reconnaitre l'action qui se déroule dans la séquence vidéo. Les expérimentations de ce système sur la base d’actions humaines de Wiezmann procurent des résultats assez intéressants comparés à d’autres méthodes plus complexes. Ces expériences montrent d'une part, la sensibilité du système pour chaque point de vue et son efficacité à reconnaitre les différentes actions, avec un taux de reconnaissance variable mais satisfaisant, ainsi que les résultats obtenus par la fusion de ces deux points de vue, qui permet l'obtention de taux de reconnaissance très performant.

The work done in this master's thesis, presents a new system for the recognition of human actions from a video sequence. The system uses, as input, a video sequence taken by a static camera. A binary segmentation method of the the video sequence is first achieved, by a learning algorithm, in order to detect and extract the different people from the background. To recognize an action, the system then exploits a set of prototypes generated from an MDS-based dimensionality reduction technique, from two different points of view in the video sequence. This dimensionality reduction technique, according to two different viewpoints, allows us to model each human action of the training base with a set of prototypes (supposed to be similar for each class) represented in a low dimensional non-linear space. The prototypes, extracted according to the two viewpoints, are fed to a $K$-NN classifier which allows us to identify the human action that takes place in the video sequence. The experiments of our model conducted on the Weizmann dataset of human actions provide interesting results compared to the other state-of-the art (and often more complicated) methods. These experiments show first the sensitivity of our model for each viewpoint and its effectiveness to recognize the different actions, with a variable but satisfactory recognition rate and also the results obtained by the fusion of these two points of view, which allows us to achieve a high performance recognition rate. / Le travail mené dans le cadre de ce projet de maîtrise vise à présenter un nouveau système de reconnaissance d’actions humaines à partir d'une séquence d'images vidéo. Le système utilise en entrée une séquence vidéo prise par une caméra statique. Une méthode de segmentation binaire est d'abord effectuée, grâce à un algorithme d’apprentissage, afin de détecter les différentes personnes de l'arrière-plan. Afin de reconnaitre une action, le système exploite ensuite un ensemble de prototypes générés, par une technique de réduction de dimensionnalité MDS, à partir de deux points de vue différents dans la séquence d'images. Cette étape de réduction de dimensionnalité, selon deux points de vue différents, permet de modéliser chaque action de la base d'apprentissage par un ensemble de prototypes (censé être relativement similaire pour chaque classe) représentés dans un espace de faible dimension non linéaire. Les prototypes extraits selon les deux points de vue sont amenés à un classifieur K-ppv qui permet de reconnaitre l'action qui se déroule dans la séquence vidéo. Les expérimentations de ce système sur la base d’actions humaines de Wiezmann procurent des résultats assez intéressants comparés à d’autres méthodes plus complexes. Ces expériences montrent d'une part, la sensibilité du système pour chaque point de vue et son efficacité à reconnaitre les différentes actions, avec un taux de reconnaissance variable mais satisfaisant, ainsi que les résultats obtenus par la fusion de ces deux points de vue, qui permet l'obtention de taux de reconnaissance très performant.