Le flot optique est un outil, prometteur et puissant, pour estimer le mouvement des objets de différentes natures, solides ou fluides. Il permet d’extraire les champs de vitesse à partir d’une séquence d’images. Dans cette étude, nous développons la méthode du flot optique pour récupérer, d’une manière précise, le champ de vitesse des mouvements turbulents incompressibles. L’estimation de turbulence consiste à minimiser une fonction d’énergie composée par un terme d’observation et un terme de régularisation. L’équation de transport d’un scalaire passif est alors employée pour représenter le terme d’observation. Cependant, dans le cas où le nombre de Reynolds est grand, et, à cause des contraintes optiques, l’image n’est pas pleinement résolue pour prendre en compte la physique de toutes les échelles de la turbulence. Pour compléter les informations manquantes liées aux physiques des petites échelles, nous adoptons une démarche similaire à celle de Large Eddy Simulation (LES), et, proposons d’utiliser le modèle mixte afin de tenir compte de l’interaction entre les grandes échelles et celles non-résolues. Quant au terme de régularisation, il se repose sur l’équation de continuité des fluides incompressibles. Les tests à l’aide des images synthétiques et expérimentales de la turbulence bi-dimensionnelle - des données des cas test de la communauté du flot optique -, ont non seulement validé notre démarche, mais montrent une amélioration significative des qualités des champs de vitesses extraites. Le cas du flot optique, en 3D, relève encore du défi dans le cas de l’estimation des champs de vitesse de la turbulence. D’une part, contrairement au 2D où il existe des cas tests bien établis, il n’existe pas, à notre connaissance, des séquences d’images 3D référentielles permettant de tester notre démarche et méthode. D’autre part, l’augmentation du coût d’estimation demande des algorithme adaptés. Ainsi, nous sommes amené à utiliser la simulation numérique directe d’écoulement turbulent en présence d’un scalaire passif, pour générer des données de scalaires afin d’évaluer la performance du flot optique. Nous prêtons également attention à l’effet du nombre de Schmidt qui caractérise la relation entre la diffusion moléculaire scalaire et la dissipation de turbulence. Les tests sont ensuite effectués avec cette base de données numériques. Les résultats montrent que la précision de l’estimation augmente avec des nombres de Schmidt plus élevés. Par ailleurs, l’influence du terme de régularisation est aussi étudié au travers deux équations qui se différencient par l’ordre spatial des dérivées partielles. Les résultats numériques montrent que l’équation avec un terme de régularisation de seconde-ordre est meilleure que celle de premier-ordre. / The method of optical flow is a powerful tool for motion estimation. It is able to extract the dense velocity field from image sequence. In this study, we employ this method to retrieve precisely the incompressible turbulent motions. For 2D turbulence estimation, it consists in minimizing an objective function constituted by an observation term and a regularization one. The observation term is based on the transport equation of a passive scalar field. For non-fully resolved scalar images, we propose to use the mixed model in large eddy simulation (LES) to determine the interaction between large-scale motions and the unresolved ones. The regularization term is based on the continuity equation of 2D incompressible flows. Evaluation of the proposed formulation is done over synthetic and experimental images. In addition, we extend optical flow to three dimensional and multiple scalar databases are generated with direct numerical simulation (DNS) in order to evaluate the performance of optical flow in the 3D context. We propose two formulations differing by the order of the regularizer. Numerical results show that the formulation with second-order regularizer outperforms its first-order counterpart. We also draw special attention to the effect of Schmidt number, which characterizes the ratio between the molecular diffusion of the scalar and the dissipation of the turbulence. Results show that the precision of the estimation increases as the Schmidt number increases. Overall, optical flow has showcased its capability of reconstructing the turbulent flow with excellent accuracy. This method has all the potential and attributes to become an effective flow measurement approach in fluid mechanics community.

Este trabalho apresenta uma nova metodologia para elastografia virtual em imagens simuladas de ultrassom utilizando métodos numéricos e métodos de visão computacional. O objetivo é estimar o módulo de elasticidade de diferentes tecidos tendo como entrada duas imagens da mesma seção transversal obtidas em instantes de tempo e pressões aplicadas diferentes. Esta metodologia consiste em calcular um campo de deslocamento das imagens com um método de fluxo óptico e aplicar um método iterativo para estimar os módulos de elasticidade (análise inversa) utilizando métodos numéricos. Para o cálculo dos deslocamentos, duas formulações são utilizadas para fluxo óptico: Lucas-Kanade e Brox. A análise inversa é realizada utilizando duas técnicas numéricas distintas: o Método dos Elementos Finitos (MEF) e o Método dos Elementos de Contorno (MEC), sendo ambos implementados em Unidades de Processamento Gráfico de uso geral, GpGPUs ( \"General Purpose Graphics Units\" ). Considerando uma quantidade qualquer de materiais a serem determinados, para a implementação do Método dos Elementos de Contorno é empregada a técnica de sub-regiões para acoplar as matrizes de diferentes estruturas identificadas na imagem. O processo de otimização utilizado para determinar as constantes elásticas é realizado de forma semi-analítica utilizando cálculo por variáveis complexas. A metodologia é testada em três etapas distintas, com simulações sem ruído, simulações com adição de ruído branco gaussiano e phantoms matemáticos utilizando rastreamento de ruído speckle. Os resultados das simulações apontam o uso do MEF como mais preciso, porém computacionalmente mais caro, enquanto o MEC apresenta erros toleráveis e maior velocidade no tempo de processamento. / This thesis presents a new methodology for computational elastography applied to simulated ultrasound images, using numerical methods and comptuter vision methods. The aim is to estimate the elastic moduli of diferent tissues using two diferent images of the same cross section acquired in diferent times and pressure conditions. The proposed methodology consists in evaluate the displacement field using optical flow techniques and then apply an inverse analysis using a numerical method. In order to evaluate the displacement field, two distinct formulations for optical flow are used: Lucas-Kanade and Brox. For the inverse analysis problem, the Finite Element Method and the Boundary Element Method are used, both implemented in general purpose graphic units, GpGPUs. Considering a number of materials that may be present in the images, the multiresgions boundary element method is used in order to couple diferent matrices for diferent materials. The optimization process is evaluated using complex variable method. The methodology is validated in three diferent steps: noiseless simulations; additive white gaussian noise simulations; and ultrasound mathematical phantom with speckle tracking. The results show that the Finite Element Method presents more accurate estimatives but a high computational cost, while the Boundary Element Method presents tolerable errors but a better processing time.

Dans cette thèse nous intéressons à la conception d'un robot mobile capable d’analyser le comportement et le mouvement d’une personne en environnement intérieur et encombré, par exemple le domicile d’une personne âgée. Plus précisément, notre objectif est de doter le robot des capacités de perception visuelle de la posture humaine de façon à mieux maîtriser certaines situations qui nécessitent de comprendre l’intention des personnes avec lesquelles le robot interagit, ou encore de détecter des situations à risques comme les chutes ou encore d’analyser les capacités motrices des personnes dont il a la garde. Le suivi de la posture dans un environnement dynamique et encombré relève plusieurs défis notamment l'apprentissage en continue du fond de la scène et l'extraction la silhouette qui peut être partiellement observable lorsque la personne est dans des endroits occultés. Ces difficultés rendent le suivi de la posture une tâche difficile. La majorité des méthodes existantes, supposent que la scène est statique et la personne est toujours visible en entier. Ces approches ne sont pas adaptées pour fonctionner dans des conditions réelles. Nous proposons, dans cette thèse, un nouveau système de suivi capable de suivre la posture de la personne dans ces conditions réelles. Notre approche utilise une grille d'occupation avec un modèle de Markov caché pour apprendre en continu l'évolution de la scène et d'extraire la silhouette, ensuite un algorithme de filtrage particulaire hiérarchique est utilisé pour reconstruire la posture. Nous proposons aussi un nouvel algorithme de gestion d'occlusion capable d'identifier et d'exclure les parties du corps cachées du processus de l'estimation de la pose. Finalement, nous avons proposé une base de données contenant des images RGB-D avec la vérité-terrain dans le but d'établir une nouvelle référence pour l'évaluation des systèmes de capture de mouvement dans un environnement réel avec occlusions. La vérité-terrain est obtenue à partir d'un système de capture de mouvement à base de marqueur de haute précision avec huit caméras infrarouges. L'ensemble des données est disponible en ligne. La deuxième contribution de cette thèse, est le développement d'une méthode de localisation visuelle à partir d'une caméra du type RGB-D montée sur un robot qui se déplace dans un environnement dynamique. En effet, le système de capture de mouvement que nous avons développé doit équiper un robot se déplaçant dans une scène. Ainsi, l'estimation de mouvement du robot est importante pour garantir une extraction de silhouette correcte pour le suivi. La difficulté majeure de la localisation d'une caméra dans un environnement dynamique, est que les objets mobiles de la scène induisent un mouvement supplémentaire qui génère des pixels aberrants. Ces pixels doivent être exclus du processus de l'estimation du mouvement de la caméra. Nous proposons ainsi une extension de la méthode de localisation dense basée sur le flux optique pour isoler les pixels aberrants en utilisant l'algorithme de RANSAC. / In this thesis we are interested in designing a mobile robot able to analyze the behavior and movement of a a person in indoor and cluttered environment. Our goal is to equip the robot by visual perception capabilities of the human posture to better analyze situations that require understanding of person with which the robot interacts, or detect risk situations such as falls or analyze motor skills of the person. Motion capture in a dynamic and crowded environment raises multiple challenges such as learning the background of the environment and extracting the silhouette that can be partially observable when the person is in hidden places. These difficulties make motion capture difficult. Most of existing methods assume that the scene is static and the person is always fully visible by the camera. These approaches are not able to work in such realistic conditions. In this thesis, We propose a new motion capture system capable of tracking a person in realistic world conditions. Our approach uses a 3D occupancy grid with a hidden Markov model to continuously learn the changing background of the scene and to extract silhouette of the person, then a hierarchical particle filtering algorithm is used to reconstruct the posture. We propose a novel occlusion management algorithm able to identify and discards hidden body parts of the person from process of the pose estimation. We also proposed a new database containing RGBD images with ground truth data in order to establish a new benchmark for the assessment of motion capture systems in a real environment with occlusions. The ground truth is obtained from a motion capture system based on high-precision marker with eight infrared cameras. All data is available online. The second contribution of this thesis is the development of a new visual odometry method to localize an RGB-D camera mounted on a robot moving in a dynamic environment. The major difficulty of the localization in a dynamic environment, is that mobile objects in the scene induce additional movement that generates outliers pixels. These pixels should be excluded from the camera motion estimation process in order to produce accurate and precise localization. We thus propose an extension of the dense localization method based on the optical flow method to remove outliers pixels using the RANSAC algorithm.

FundaÃÃo Cearense de Apoio ao Desenvolvimento Cientifico e TecnolÃgico / O auxÃlio ao diagnÃstico atravÃs da visualizaÃÃo de imagens mÃdicas vÃm sendo utilizado em diversas Ãreas da Medicina tais como pneumologia, cardiologia, traumatologia, neurologia, dentre outras. Na Ãrea da cardiologia, vÃrias aplicaÃÃes clÃnicas tÃm sido propostas para a anÃlise de doenÃas cardÃacas atravÃs da quantificaÃÃo e avaliaÃÃo da dessincronia ventricular esquerda durante a deformaÃÃo do mÃsculo cardÃaco (miocÃrdio). Existem, atualmente, duas tÃcnicas utilizadas na aferiÃÃo da deformaÃÃo miocÃrdica em duas dimensÃes: Doppler Tecidual (DT) e Strain 2D (St2D). A primeira tÃcnica possui desvantagens quanto à dependÃncia do Ãngulo de insonaÃÃo do transdutor durante o exame ecocardiogrÃfico, diminuindo a chance de reprodutibilidade do resultado das mediÃÃes entre especialistas. A segunda tÃcnica, recentemente introduzida e tambÃm chamada de Speckle Tracking, consiste no acompanhamento de marcadores acÃsticos naturais existentes na imagem produzida pelo ultrassom. Neste sentido, vÃ-se a importÃncia do estudo de tÃcnicas para rastrear esses marcadores acÃsticos. A presente dissertaÃÃo realiza uma anÃlise comparativa entre oito algoritmos de estimaÃÃo de deslocamento baseados na tÃcnica de Casamento de Blocos (CB) e trÃs algoritmos baseados na tÃcnica de Fluxo Ãptico (FO), que sÃo as duas atuais tÃcnicas amplamente citadas na literatura. A anÃlise à realizada mediante vÃdeos sintÃticos e vÃdeos mÃdicos de exames ecocardiogrÃficos. A avaliaÃÃo das tÃcnicas em vÃdeos sintÃticos à realizada quanto à trajetÃria e à deformaÃÃo. Jà a avaliaÃÃo em vÃdeos de exames ecocardiogrÃficos à realizada quanto Ãs curvas e taxas de deformaÃÃo. Na anÃlise da trajetÃria sÃo aplicadas duas mÃtricas de avaliaÃÃo das tÃcnicas: correlaÃÃo mÃdia e erro quadrÃtico mÃdio. Para a anÃlise das curvas e das taxas de deformaÃÃo a mÃtrica usada à o valor do erro quadrÃtico mÃdio em relaÃÃo à deformaÃÃo global (global strain) do miocÃrdio. Os resultados indicam que o desempenho idÃntico de alguns estimadores de deslocamento os reduzem de oito para seis algoritmos. A tÃcnica de CB mostra-se viÃvel para o rastreamento de marcas acÃsticas mas à dependente das dimensÃes adotadas nos blocos. Em relaÃÃo Ãs tÃcnicas de FO, o algoritmo de Lucas e Kanade Piramidal à o que obtÃm melhor resultado nos testes realizados, produzindo curvas de deformaÃÃo global com erro mÃdio de 0,47%, enquanto os valores de erro dos outros algoritmos de FO estÃo em torno de 10%. No caso, os erros dos estimadores de CB variam de 1% a 16%. / Aided diagnosis by visualization of medical images has been used in several medical fields such as pulmonology, cardiology, traumatology, neurology, and others. In cardiology, several clinical applications have been proposed for the analysis of heart disease by quantification and evaluation of ventricular dyssynchrony during deformation of the heart muscle (myocardium). There are currently two techniques used in the measurement of myocardial deformation in two dimensions: Tissue Doppler and 2D Strain. The first technique has drawbacks regarding the dependence on the angle of insonation of the transducer during the echocardiographic examination, which reduce the chance of reproducibility of measurements among experts. The second technique, recently introduced and also called Speckle Tracking, consists of tracking the natural acoustic markers in the image produced by ultrasound. In this sense we see the importance of studying techniques to track these acoustic markers. This thesis performs a comparative analysis of eight algorithms from time-delay estimators based on the block matching technique and three algorithms based on the optical flow technique, which are the two current techniques widely presented in the literature. The analysis is performed using synthetic videos and medical videos from echocardiographic examinations. The evaluation of the techniques in synthetic videos is performed on the trajectory and deformation. The assessment in echocardiographic videos is held regarding the strain curves and strain rates. In the analysis of the trajectory are applied two metrics for evaluating techniques: mean correlation and mean square error. For the analysis of strain curves and of strain rate the measure used is the value of the mean square error relative to global strain of myocardium. The results indicate that the identical performance of some estimators reduce the time-delay estimators from eight to six algorithms. The block matching technique appears to be a viable technique for tracking acoustic marks but is dependent on the dimensions adopted in the blocks. Regarding optical flow techniques, the Lucas and Kanade Pyramidal algorithm is the one which gets the best results in the tests performed herein and produce global strain curves average error of 0.47 %, while the error values of the other optical flow algorithms are around 10 %. In case, the block matching time-delay estimators errors vary from 1% to 16%.

Este trabalho apresenta uma nova metodologia para elastografia virtual em imagens simuladas de ultrassom utilizando métodos numéricos e métodos de visão computacional. O objetivo é estimar o módulo de elasticidade de diferentes tecidos tendo como entrada duas imagens da mesma seção transversal obtidas em instantes de tempo e pressões aplicadas diferentes. Esta metodologia consiste em calcular um campo de deslocamento das imagens com um método de fluxo óptico e aplicar um método iterativo para estimar os módulos de elasticidade (análise inversa) utilizando métodos numéricos. Para o cálculo dos deslocamentos, duas formulações são utilizadas para fluxo óptico: Lucas-Kanade e Brox. A análise inversa é realizada utilizando duas técnicas numéricas distintas: o Método dos Elementos Finitos (MEF) e o Método dos Elementos de Contorno (MEC), sendo ambos implementados em Unidades de Processamento Gráfico de uso geral, GpGPUs ( \"General Purpose Graphics Units\" ). Considerando uma quantidade qualquer de materiais a serem determinados, para a implementação do Método dos Elementos de Contorno é empregada a técnica de sub-regiões para acoplar as matrizes de diferentes estruturas identificadas na imagem. O processo de otimização utilizado para determinar as constantes elásticas é realizado de forma semi-analítica utilizando cálculo por variáveis complexas. A metodologia é testada em três etapas distintas, com simulações sem ruído, simulações com adição de ruído branco gaussiano e phantoms matemáticos utilizando rastreamento de ruído speckle. Os resultados das simulações apontam o uso do MEF como mais preciso, porém computacionalmente mais caro, enquanto o MEC apresenta erros toleráveis e maior velocidade no tempo de processamento. / This thesis presents a new methodology for computational elastography applied to simulated ultrasound images, using numerical methods and comptuter vision methods. The aim is to estimate the elastic moduli of diferent tissues using two diferent images of the same cross section acquired in diferent times and pressure conditions. The proposed methodology consists in evaluate the displacement field using optical flow techniques and then apply an inverse analysis using a numerical method. In order to evaluate the displacement field, two distinct formulations for optical flow are used: Lucas-Kanade and Brox. For the inverse analysis problem, the Finite Element Method and the Boundary Element Method are used, both implemented in general purpose graphic units, GpGPUs. Considering a number of materials that may be present in the images, the multiresgions boundary element method is used in order to couple diferent matrices for diferent materials. The optimization process is evaluated using complex variable method. The methodology is validated in three diferent steps: noiseless simulations; additive white gaussian noise simulations; and ultrasound mathematical phantom with speckle tracking. The results show that the Finite Element Method presents more accurate estimatives but a high computational cost, while the Boundary Element Method presents tolerable errors but a better processing time.

Cette thèse étudie différentes relaxations pour minimiser des fonctionnelles non convexes qui apparaissent en traitement d’images. Des problèmes comme la segmentation d’image peuvent en effet s’écrire comme un problème de minimisation d’une certaine fonctionnelle, le minimiseur représentant la segmentation recherchée. Différentes méthodes ont été proposées pour trouver des minima locaux ou globaux de la fonctionnelle non convexe du modèle de Mumford-Shah constant par morceaux à deux phases. Certaines approches utilisent une relaxation convexe qui permet d’obtenir des minima globaux de la fonctionnelle non convexe. On rappelle et compare certaines de ces méthodes et on propose un nouveau modèle par bande étroite, qui permet d’obtenir des minima locaux tout en utilisant des algorithmes robustes qui proviennent de l’optimisation convexe. Ensuite, on construit une relaxation convexe d’un modèle de segmentation à deux phases qui repose sur la comparaison entre deux histogrammes donnés et les histogrammes estimés globalement sur les deux régions de la segmentation. Des relaxations pour des problèmes multi-étiquettes à plusieurs dimensions comme le flot optique sont également étudiées. On propose une relaxation convexe avec un algorithme itératif qui ne comprend que des projections qui se calculent exactement, ainsi qu’un nouvel algorithme pour une relaxation convexe sur chaque variable mais non convexe globalement. On étudie la manière d’estimer une solution du problème non convexe original à partir d’une solution d’un problème relaxé en comparant des méthodes existantes avec des nouvelles / In this thesis we study different relaxations of non-convex functionals that can be found in image processing. Some problems, such as image segmentation, can indeed be written as the minimization of a functional. The minimizer of the functional represents the segmentation. Different methods have been proposed in order to find local or global minima of the non-convex functional of the two-phase piecewise constant Mumford-Shah model. With a convex relaxation of this model we can find a global minimum of the nonconvex functional. We present and compare some of these methods and we propose a new model with a narrow band. This model finds local minima while using robust convex optimization algorithms. Then a convex relaxation of a two-phase segmentation model is built that compares two given histograms with those of the two segmented regions. We also study some relaxations of high-dimension multi-label problems such as optical flow computation. A convex relaxation with a new algorithm is proposed. The algorithm is iterative with exact projections. A new algorithm is given for a relaxationthat is convex in each variable but that is not convex globally. We study the problem of constructing a solution of the original non-convex problem with a solution of the relaxed problem. We compare existing methods with new ones.

Flugzeug- und satellitengestützte Zeilenkameras ermöglichen eine sehr ökonomische Aufnahme von hoch aufgelösten Luftbildern mit großer Schwadbreite. Eine ungleichförmige Bewegung der Kamera kann sich auf die Bildqualität auswirken. Deswegen ist es unerlässlich, schnelle Orientierungsänderungen der Kamera mit angemessener Genauigkeit und Messrate zu erfassen. Deshalb ist es unerlässlich, die Orientierung der Kamera genau zu messen, um sicher zu stellen, dass die resultierenden Bilder in einem Nachbearbeitungsschritt geometrisch korrigiert werden können. Angemessene High-End-Navigationssysteme sind groß und teuer und ihre Genauigkeit und Messrate dennoch für viele denkbare Anwendungen unzureichend. Aus diesen Gründen besteht ein großes Interesse an Methoden zur Unterstützung der Orientierungsmessung durch die Nutzung optischer Informationen vom Hauptobjektiv bzw. Teleskop. In dieser Arbeit werden zwei unterschiedliche Verfahren vorgestellt, die es erlauben, schnelle Orientierungsänderungen der Kamera auf optischem Wege zu ermitteln. Ersteres basiert auf zusätzlichen Bildsensoren mit kleiner Fläche. Der optische Fluss auf diesen Bildsensoren wird ermittelt und zur Bestimmung von Orientierungsänderungen beliebiger Fernerkundungssysteme verwendet. Das zweite Verfahren beruht ausschließlich auf den Inhalten der Zeilenbilder und der gemessenen Kameratrajektorie. Hierfür macht sich das Verfahren die typische Geometrie multispektraler Zeilenkameras zu Nutze. Zunächst werden homologe Punkte in den möglicherweise stark verzerrten Zeilenbildern unterschiedlicher Spektralbänder extrahiert. Diese Punkte werden dann dazu benutzt, die Orientierungsänderungen der Kamera zu ermitteln. Schließlich wird gezeigt, dass es möglich ist, die absolute Orientierung einer luftgestützten Zeilenkamera anhand der optisch ermittelten Orientierungsänderungen hochgenau zu ermitteln. / Airborne and spaceborne line cameras allow a very economic acquisition of high resolution and wide swath images of the Earth. They generate two-dimensional images while rotating or translating, which causes a non-rigid image geometry. Nonuniform motion of the camera can negatively affect the image quality. Due to this, a key requirement for line cameras is that fast orientation changes have to be measured with a very high rate and precision. Therefore, it is essential to measure the camera’s orientation accurately to ensure that the resulting imagery products can be geometrically corrected in a later processing step. Adequate high-end measurement systems are large and expensive and their angular and temporal resolution can be still too low for many possible applications. Due to these reasons there is great interest in approaches to support the orientation measurement by using optical information received through the main optics or telescope. In this thesis two different approaches are presented that allow the determination of a line camera’s orientation changes optically. One approach to determine fast orientation changes is based on small auxiliary frame image sensors. The optical flow on these sensors is determined and used to derive orientation changes of any remote sensing system. The second approach does not require any additional sensors to determine orientation changes. It relies only on the images of a line camera and a rough estimate of its trajectory by taking advantage of the typical geometry of multi-spectral line cameras. In a first step homologous points are detected within the distorted images of different spectral bands. These points are used to calculate the orientation changes of the camera with a high temporal and angular resolution via bundle adjustment. Finally it is shown how the absolute exterior orientation of an airborne line camera can completely be derived from the optically determined orientation changes.

<p>The problem of separating a non-rectangular foreground image from a background image is a classical problem in image processing and analysis, known as matting or keying. A common example is a film frame where an actor is extracted from the background to later be placed on a different background. Compositing of these objects against a new background is one of the most common operations in the creation of visual effects. When the original background is of non-constant color the matting becomes an under determined problem, for which a unique solution cannot be found. </p><p>This thesis describes a framework for computing mattes from images with backgrounds of non-constant color, using Bayesian statistics. Foreground and background color distributions are modeled as oriented Gaussians and optimal color and opacity values are determined using a maximum a posteriori approach. Together with information from optical flow algorithms, the framework produces mattes for image sequences without needing user input for each frame. </p><p>The approach used in this thesis differs from previous research in a few areas. The optimal order of processing is determined in a different way and sampling of color values is changed to work more efficiently on high-resolution images. Finally a gradient-guided local smoothness constraint can optionally be used to improve results for cases where the normal technique produces poor results.</p>

Optical flow is used in a number of computer vision applications. However, its use in mobile robotic applications is limited because of the high computational complexity involved and the limited availability of computational resources on such platforms. The lack of a hardware that is capable of computing optical flow vector field in real time is a factor that prevents the mobile robotics community to efficiently utilize some successful techniques presented in computer vision literature. In this thesis work, we design and implement a high performance FPGA hardware with a small footprint and low power consumption that is capable of providing over-realtime optical flow data and is hence suitable for this application domain. A well known differential optical flow algorithm presented by Horn &amp / Schunck is selected for this implementation. The complete hardware design of the proposed system is described in details. We also discuss the design alternatives and the selected approaches together with a discussion of the selection procedure. We present the performance analysis of the proposed hardware in terms of computation speed, power consumption and accuracy. The designed hardware is tested with some of the available test sequences that are frequently used for performance evaluations of the optical flow techniques in literature. The proposed hardware is capable of computing optical flow vector field on 256x256 pixels images in 3.89ms which corresponds to a processing speed of 257 fps. The results obtained from FPGA implementation are compared with a floating-point implementation of the same algorithm realized on a PC hardware. The obtained results show that the hardware implementation achieved a superior performance in terms of speed, power consumption and compactness while there is minimal loss of accuracy due to the fixed point implementation.

This thesis deals with image segmentation problems that arise in various computer vision related fields such as medical imaging, satellite imaging, video surveillance, recognition and robotic vision. More specifically, this thesis deals with a special class of image segmentation technique called Snakes or Active Contour Models. In active contour models, image segmentation is posed as an energy minimization problem, where an objective energy function (based on certain image related features) is defined on the segmenting curve (contour). Typically, a gradient descent energy minimization approach is used to drive the initial contour towards a minimum for the defined energy. The drawback associated with this approach is that the contour has a tendency to get stuck at undesired local minima caused by subtle and undesired image features/edges. Thus, active contour based curve evolution approaches are very sensitive to initialization and noise. The central theme of this thesis is to develop techniques that can make active contour models robust against certain classes of local minima by incorporating global information in energy minimization. These techniques lead to energy minimization with global considerations; we call these models -- 'Non-local active contours'. In this thesis, we consider three widely used active contour models: 1) Edge- and region-based segmentation model, 2) Prior shape knowledge based segmentation model, and 3) Motion segmentation model. We analyze the traditional techniques used for these models and establish the need for robust models that avoid local minima. We address the local minima problem for each model by adding global image considerations.