The limitations of foreground segmentation in difficult environments using standard color space features often result in poor performance during autonomous tracking. This work presents a new approach for classification of foreground and background pixels in image sequences by employing an ensemble of classifiers, each operating on a different feature type such as the three RGB features, gradient magnitude and orientation features, and eight Haar features. These thirteen features are used in an ensemble classifier where each classifier operates on a single image feature. Each classifier implements a Mixture of Gaussians-based unsupervised background classification algorithm. The non-thresholded, classification decision score of each classifier are fused together by taking the average of their outputs and creating one single hypothesis. The results of using the ensemble classifier on three separate and distinct data sets are compared to using only RGB features through ROC graphs. The extended feature vector outperforms the RGB features on all three data sets, and shows a large scale improvement on two of the three data sets. The two data sets with the greatest improvements are both outdoor data sets with global illumination changes and the other has many local illumination changes. When using the entire feature set, to operate at a 90% true positive rate, the per pixel, false alarm rate is reduced five times in one data set and six times in the other data set.

The main object of study of this work are orbits in so-called minimal sheets in irreducible representations of semisimple groups. Let $G$ be a semisimple group. The notion of sheets goes back to Dixmier: Given a $G$-module $V$, the union of all orbits in $V$ of a fixed dimension is a locally closed subset. Its irreducible components are called sheets of $V$. We call a sheet minimal if it contains an orbit in $V$ of minimal strictly positive dimension among all orbits in $V$. In Chapter I, some notation is fixed and some basic results are proved. In Chapter II, we describe minimal sheets in simple $G$-modules, and study $G$-stable deformations of orbits in minimal sheets by means of an invariant Hilbert scheme. Invariant Hilbert Schemes have been introduced by Alexeev and Brion in 2005. These are quasi-projective schemes representing functors of families of $G$-schemes with prescribed Hilbert function. The discussion in Chapter II is closely related to the work of Jansou in the following way: Choose once and for all a highest weight vector $v_\lambda \in V(\lambda)$ for each dominant weight $\lambda \in \Lambda^+$, and let $X_\lambda = \overline{G v_\lambda} \subset V(\lambda)$ be the closure of the orbit $G v_\lambda$ of $v_\lambda$ in $V(\lambda)$. In his thesis Jansou investigates $G$-stable deformations of $X_\lambda$ in $V(\lambda)$. If $h_\lambda$ denotes the Hilbert function of $X_\lambda$, then Jansou proves that the invariant Hilbert scheme $Hilb^G_{h_\lambda}(V(\lambda))$ is an affine space of dimension 0 or 1, depending on $G$ and $\lambda$. Furthermore, he gives a complete list of all pairs $(G,\lambda)$ such that $Hilb^G_{h_\lambda}(V(\lambda))$ is an affine line. In the sequel, we call these weights Jansou-weights. The orbit $Gv_\lambda$ is of minimal strictly positive dimension among all $G$-orbits in $V(\lambda)$. There exist other orbit of the same dimension as $Gv_\lambda$ in $V(\lambda)$ if and only if $\lambda$ is an integral multiple of a Jansou-weight. Here, we start with a general orbit $X$ of minimal strictly positive dimension in a fixed simple $G$-module $V(\lambda)$, and we study $G$-stable deformations of $X$. In particular, we conjecture that the invariant Hilbert scheme parametrizing the $G$-stable deformations of $X$ in the closure of the sheet of $X$ is an affine space of dimension either 0 or 1. This will stand in contrast to the fact that the invariant Hilbert scheme parametrizing the $G$-stable deformations of $X$ in $V(\lambda)$ can look much more complicated. This is the content of Chapter III, in which we will focus on the group $\SL_2$, and compute some corresponding invariant Hilbert schemes. In particular, we study deformations of orbits of the form $SL_2 \cdot x^{d/2}y^{d/2}$ in the space $k[x,y]_d = V(d)$ of binary forms of degree $d$. It turns out that easiest accessible case is when $d$ is a multiple of 4, and even in this case the corresponding invariant Hilbert scheme can become very complicated. This reflects the principle that even in `simple' cases for invariant Hilbert schemes all possible sort of `bad' things (different irreducible components, non-reduced points, singularities) occur. (This `bad' behavior is also encountered in the case of the classical Grothendieck Hilbert scheme parametrizing closed subschemes of projective space with a given Hilbert polynomial.) In Chapter III Classical Invariant Theory is often used, and some computations are computer-based. Finally, in Chapter IV we turn our attention to not necessarily simple modules. In the multiplicity-free case important work has been done by Bravi and Cupit-Foutou. We translate some of their results to the case of not necessarily multiplicity-free modules. This corrects a result by Alexeev and Brion. Chapter IV is independent from the preceding chapters.

<p>The aim of this work has been to develop new knowledge about macromodelling of microsystems. Doing that, we have followed two different approaches for generating macromodels, namely model order reduction and lumped modelling. The latter is a rather mature method that has been widely recognized and used for a relatively long period of time. Model order reduction, on the other hand, is a relatively new area still in rapid development. Due to this, the parts considering reduced order modelling is strongly biased towards methodology and concepts, whereas parts on lumped modelling are biased towards systems and devices.</p><p>In the first part of this thesis, we focus on model order reduction. We introduce some approaches for reducing model order for linear systems, and we give an example related to squeeze-film damping. We then move on to investigate model order reduction of nonlinear systems, where we present and use the concept of invariant manifolds. While the concept of invariant manifolds is general, we utilize it for reducing models. An obvious advantage of using invariant manifold theory is that it offers a conceptually clear understanding of effects and behaviour of nonlinear system.</p><p>We exemplify and investigate the accuracy of one method for identifying invariant manifolds. The example is based on an industrialized dual-axis accelerometer.</p><p>A new geometrical interpretation of external forcing, relating to invariant manifolds, is presented. We show how this can be utilized to deal with external forcing in a manner consistent with the invariance property of the manifold. The interpretation also aids in reducing errors for reduce models.</p><p>We extend the asymptotic approach in a manner that makes it possible to create design-parameter sensitive models. We investigate an industrialized dual-axis accelerometer by means of the method and demonstrate capabilities of the method. We also discuss how manifolds for nonlinear dissipative systems can be found.</p><p>Focusing on lumped modelling, we analyse a microresonator. We also discuss the two analogies that can be used to build electrical equivalents of mechanical systems. It is shown how the f → V analogy, linking velocity to voltage, is the natural choice. General properties of lumped modelling are investigated using models with varying degrees of freedom.</p><p>Finally, we analyse an electromagnetic system, intended for levitating objects, and we demonstrate the scaling effects of the system. Furthermore, we prove the intrinsic stability of the system, although the floating disc will be slightly tilted. This is the first analysis done assessing the stability criterions of such a systems. The knowledge arising from the analysis gives strong indications on how such a system can be utilized, designed, and improved.</p>

Thesis (Ph. D.)--University of Rhode Island, 2005. / Typescript. Includes bibliographical references (leaves 87-89).

No description available.

We introduce a natural symmetry condition for a pseudodifferential operator on a manifold with cylindrical ends ensuring that the operator admits a doubling across the boundary. For such operators we prove an explicit index formula containing, apart from the Atiyah-Singer integral, a finite number of residues of the logarithmic derivative of the conormal symbol.

The aim of this work has been to develop new knowledge about macromodelling of microsystems. Doing that, we have followed two different approaches for generating macromodels, namely model order reduction and lumped modelling. The latter is a rather mature method that has been widely recognized and used for a relatively long period of time. Model order reduction, on the other hand, is a relatively new area still in rapid development. Due to this, the parts considering reduced order modelling is strongly biased towards methodology and concepts, whereas parts on lumped modelling are biased towards systems and devices. In the first part of this thesis, we focus on model order reduction. We introduce some approaches for reducing model order for linear systems, and we give an example related to squeeze-film damping. We then move on to investigate model order reduction of nonlinear systems, where we present and use the concept of invariant manifolds. While the concept of invariant manifolds is general, we utilize it for reducing models. An obvious advantage of using invariant manifold theory is that it offers a conceptually clear understanding of effects and behaviour of nonlinear system. We exemplify and investigate the accuracy of one method for identifying invariant manifolds. The example is based on an industrialized dual-axis accelerometer. A new geometrical interpretation of external forcing, relating to invariant manifolds, is presented. We show how this can be utilized to deal with external forcing in a manner consistent with the invariance property of the manifold. The interpretation also aids in reducing errors for reduce models. We extend the asymptotic approach in a manner that makes it possible to create design-parameter sensitive models. We investigate an industrialized dual-axis accelerometer by means of the method and demonstrate capabilities of the method. We also discuss how manifolds for nonlinear dissipative systems can be found. Focusing on lumped modelling, we analyse a microresonator. We also discuss the two analogies that can be used to build electrical equivalents of mechanical systems. It is shown how the f → V analogy, linking velocity to voltage, is the natural choice. General properties of lumped modelling are investigated using models with varying degrees of freedom. Finally, we analyse an electromagnetic system, intended for levitating objects, and we demonstrate the scaling effects of the system. Furthermore, we prove the intrinsic stability of the system, although the floating disc will be slightly tilted. This is the first analysis done assessing the stability criterions of such a systems. The knowledge arising from the analysis gives strong indications on how such a system can be utilized, designed, and improved.

L'objectiu bàsic d'aquesta tesi és l'estudi de la dinàmica a l'entorn de tors de dimensió baixa de sistemes hamiltonians analítics. Per aquest estudi l'eina fonamental és l'ús de formes normals al voltant d'aquests objectes.La formulació dels resultats d'aquesta memòria s'ha fet de manera adient per a la seva aplicació a models de mecànica celeste del món real. Per aquest motiu els resultats no es redueixen només al cas autònom, sinó que algun d'ells pren en consideració la possiblitat que les diferents perturbacions pugin dependre del temps de forma periòdica o quasiperiòdica. Aquests resultats s'apliquen per descriure la dinàmica d'alguns problemes d'interes per la Astronàutica. Per tant, els resultats obtinguts inclouen també aplicacions numèriques.Els resultats assolits en cadascun del capítols de la memòria es poden sintetitzar de la forma següent:Capítol 1.- Estudi de la dinàmica entorn d'un tor parcialment el.líptic d'un sistema Hamiltonià autònom. Es donen cotes inferiors pel temps de difusió entorn d'un tor totalment el.líptic, així com estimacions, en el cas general, de la densitat de tors invariants (de qualsevol dimensió) al voltant del tor inicial. Les estimacions en la velocitat de difusió i en la proximitat a 1 d'aquesta densitat, són exponencialment petites respecte la distància al tor inicial.Capítol 2.- Computació numèrica de formes normals al voltant d'òrbites periòdiques. Es desenvolupa un mètode per a calcular formes normals al voltant d'òrbites periòdiques el.líptiques de sistemes hamiltonians. Aquesta metodologia és aplicada numèricament a una òrbita periòdica del Problema Restringit de tres Cossos espaial. Els resultats d'aquest capítol es poden veure com una implementació numèrica del Capítol 1.Capítol 3.- Persistència de tors de dimensió baixa sota perturbacions quasiperiòdiques. Es mostra que un tor de dimensió baixa d'un sistema hamiltonià sotmès a una perturbació quasiperiòdica és pot continuar respecte el paràmetre perturbatiu, tot afegint a les freqüències bàsiques inicials les de la perturbació, excepte per un conjunt de mesura petita pel paràmetre. Al igual que en el Capítol 1 també s'estima la densitat de tors en el problema perturbat. En ambdós casos, les cotes obtingudes per la mesura dels tors pels qual no és possible provar existència són de tipus exponencialment petit.Apèndix. Es presenta un resultat obtingut de forma conjunta amb Rafael Ramírez-Ros sobre la reducció a coeficients constants de sistemes d'equacions lineals autònoms perturbats quasiperiòdicament. Es mostra que tal reducció és possible excepte un reste exponencialment petit en el tamany de la perturbació.

Nous étudions les poids de graphes (c'est-à-dire, les invariants de graphes) qui apparaissent naturellement dans la théorie de Mayer et la théorie de Ree-Hoover pour le développement du viriel dans le contexte d'un gaz imparfait. Nous portons une attention particulière au deuxième poids ωM(C) de Mayer et au poids ωRH(C) de Ree-Hoover d'un graphe 2-connexe c dans le cas d'un gaz à noyaux durs et à positions continues en une dimension. Ces poids sont calculés à partir de volumes signés de polytopes convexes associés au graphe c en utilisant la méthode des homomorphismes de graphes, que nous avons aussi adaptée au cas du poids de Ree-Hoover, ainsi que les transformées de Fourier. En faisant appel à l'inversion de Möbius, nous présentons des relations entre les poids de Mayer et de Ree-Hoover. Ces relations nous permettent de donner une définition simple explicite du concept du "star content" introduit par Ree-Hoover et d'analyser certaines de ses propriétés fondamentales. Parmi nos résultats, nous donnons des tables contenant les valeurs du poids de Mayer et du poids de Ree-Hoover pour tous les graphes 2-connexes de taille au plus 8 ainsi que d'autres paramètres descriptifs. Nous développons aussi des formules explicites pour les poids de Mayer et de Ree-Hoover pour certaines familles de graphes 2-connexes simplement, doublement et triplement infinies, incluant par exemple, le poids de Mayer des graphes bipartis complets K m,n. En analysant les tables précédentes à l'aide du logiciel Maple, nous montrons que les poids de Mayer et de Ree-Hoover ne sont pas exprimables comme des fonctions faisant seulement appel à certains paramètres classiques de la théorie des graphes. Finalement, nous présentons une méthode générale pour le calcul du poids de Mayer d'un graphe connexe quelconque basée sur les arborescences couvrantes en utilisant les transformées de Fourier. Nous illustrons cette méthode sur des cas particuliers incluant les particules dures en dimension quelconque d. Cette méthode donne aussi lieu à un algorithme de calcul basé sur les différences divisées pour le cas des particules dures en dimension d = 1. ______________________________________________________________________________ MOTS-CLÉS DE L’AUTEUR : Poids de Mayer, Poids de Ree-Hoover, Mécanique statistique, Méthode des homomorphismes de graphes, Transformées de Fourier, Gaz imparfaits.

This is an expository thesis which addresses the requirements for an operator algebra to be similar to a <em>C</em>*-algebra. It has been conjectured that this similarity condition is equivalent to either amenability or total reductivity; however, the problem has only been solved for specific types of operators. <br /><br /> We define amenability and total reductivity, as well as present some of the implications of these properties. For the purpose of establishing the desired result in specific cases, we describe the properties of two well-known types of operators, namely the compact operators and quasitriangular operators. Finally, we show that if A is an algebra of compact operators or of triangular operators then A is similar to a <em>C</em>* algebra if and only if it has the total reduction property.