Maximum likelihood is an ubiquitous method used in the estimation of generalized linear mixed model (GLMM). However, the method entails computational difficulties and relies on the normality assumption for random effects. We propose a second-order least squares (SLS) estimator based on the first two marginal moments of the response variables. The proposed estimator is computationally feasible and requires less distributional assumptions than the maximum likelihood estimator. To overcome the numerical difficulties of minimizing an objective function that involves multiple integrals, a simulation-based SLS estimator is proposed. We show that the SLS estimators are consistent and asymptotically normally distributed under fairly general conditions in the framework of GLMM. Missing data is almost inevitable in longitudinal studies. Problems arise if the missing data mechanism is related to the response process. This thesis develops the proposed estimators to deal with response data missing at random by either adapting the inverse probability weight method or applying the multiple imputation approach. In practice, some of the covariates are not directly observed but are measured with error. It is well-known that simply substituting a proxy variable for the unobserved covariate in the model will generally lead to biased and inconsistent estimates. We propose the instrumental variable method for the consistent estimation of GLMM with covariate measurement error. The proposed approach does not need any parametric assumption on the distribution of the unknown covariates. This makes the method less restrictive than other methods that rely on either a parametric distribution of the covariates, or to estimate the distribution using some extra information. In the presence of data outliers, it is a concern that the SLS estimators may be vulnerable due to the second-order moments. We investigated the robustness property of the SLS estimators using their influence functions. We showed that the proposed estimators have a bounded influence function and a redescending property so they are robust to outliers. The finite sample performance and property of the SLS estimators are studied and compared with other popular estimators in the literature through simulation studies and real world data examples.

This thesis focuses on two statistical problems related to credit scoring. In credit scoring of individuals, two classes are distinguished, namely low and high risk individuals (the so-called "good" and "bad" risk classes). Firstly, we suggest a measure which may be used to study the nature of a classifier for distinguishing between the two risk classes. Secondly, we derive a new method DOUW (detecting outliers using weights) which may be used to fit logistic regression models robustly and for the detection of outliers. In the first problem, the focus is on a measure which may be used to study the nature of a classifier. This measure transforms a random variable so that it has the same distribution as another random variable. Assuming a linear form of this measure, three methods for estimating the parameters (slope and intercept) and for constructing confidence bands are developed and compared by means of a Monte Carlo study. The application of these estimators is illustrated on a number of datasets. We also construct statistical hypothesis to test this linearity assumption. In the second problem, the focus is on providing a robust logistic regression fit and the identification of outliers. It is well-known that maximum likelihood estimators of logistic regression parameters are adversely affected by outliers. We propose a robust approach that also serves as an outlier detection procedure and is called DOUW. The approach is based on associating high and low weights with the observations as a result of the likelihood maximization. It turns out that the outliers are those observations to which low weights are assigned. This procedure depends on two tuning constants. A simulation study is presented to show the effects of these constants on the performance of the proposed methodology. The results are presented in terms of four benchmark datasets as well as a large new dataset from the application area of retail marketing campaign analysis. In the last chapter we apply the techniques developed in this thesis on a practical credit scoring dataset. We show that the DOUW method improves the classifier performance and that the measure developed to study the nature of a classifier is useful in a credit scoring context and may be used for assessing whether the distribution of the good and the bad risk individuals is from the same translation-scale family. / Thesis (Ph.D. (Risk Analysis))--North-West University, Potchefstroom Campus, 2008.

Σκοπός της εργασίας αυτής είναι να εφαρμόσουμε την Παραγοντική Ανάλυση στο δείγμα μας, έτσι ώστε να ανιχνεύσουμε και να περιγράψουμε τις καταναλωτικές συνήθειες του Ελληνικού πληθυσμού ως προς την κατανάλωση 9 κατηγοριών αλκοολούχων ποτών. Η εφαρμογή της μεθόδου γίνεται με την χρήση του στατιστικού προγράμματος SPSS. Στο πρώτο κεφάλαιο παρουσιάζεται η οικογένεια μεθόδων επίλυσης του προβλήματος και στο δεύτερο η μέθοδος που επιλέχτηκε για την επίλυση, η Παραγοντική Ανάλυση. Προσδιορίζουμε το αντικείμενο, τα στάδια σχεδιασμού και τις προϋποθέσεις της μεθόδου, καθώς και τα κριτήρια αξιολόγησης των αποτελεσμάτων. Τα κεφάλαια που ακολουθούν αποτελούν το πρακτικό μέρος της εργασίας. Στο 3ο κεφάλαιο αναφέρουμε την πηγή των δεδομένων μας και την διεξαγωγή του τρόπου συλλογής τους. Ακολουθεί ο εντοπισμός των «χαμένων» απαντήσεων και εφαρμόζεται η Ανάλυση των Χαμένων Τιμών (Missing Values Analysis) για τον προσδιορισμό του είδους αυτών και την αποκατάσταση τους στο δείγμα. Στην συνέχεια παρουσιάζουμε το δείγμα μας με τη βοήθεια της περιγραφικής στατιστικής και τέλος δημιουργούμε και περιγράφουμε το τελικό μητρώο δεδομένων το οποίο θα αναλύσουμε παραγοντικά. Στο 4ο και τελευταίο κεφάλαιο διερευνάται η καταλληλότητα του δείγματος για την εφαρμογή της Παραγοντικής Ανάλυσης με τον έλεγχο της ικανοποίησης των προϋποθέσεων της μεθόδου. Ακολουθεί η παράλληλη μελέτη του δείγματος συμπεριλαμβάνοντας και μη στην επίλυση τις ακραίες τιμές (outliers) που εντοπίστηκαν. Καταλήγοντας στο συμπέρασμα ότι οι ακραίες τιμές δεν επηρεάζουν τα αποτελέσματα της μεθόδου, εφαρμόζουμε την Παραγοντική Ανάλυση με τη χρήση της μεθόδου των κυρίων συνιστωσών και αναφέρουμε αναλυτικά όλα τα βήματα μέχρι να καταλήξουμε στα τελικά συμπεράσματα μας. / The purpose of this paper is to apply the Factor Analysis to our sample in order to detect and describe patterns concerning the consumption of 9 categories of alcoholic beverages by the Greek population. For the application of the method, we use the statistical program SPSS. The first chapter presents the available methods for solving this problem and the second one presents the chosen method, namely Factor Analysis. We specify the objective of the analysis, the design and the critical assumptions of the method, as well as the criteria for the evaluation of the results. In the third chapter we present the source of our data and how the sampling was performed. Furthermore, we identify the missing values and we apply the Missing Values Analysis to determine their type. We also present our sample using descriptive statistics and then create and describe the final matrix which we analyze with Factor Analysis. In the fourth and last chapter we investigate the suitability of our samples for applying Factor Analysis. In the sequence, we perform the parallel study of our sample both including and not including the extreme values that we identified (which we call “outliers”). We conclude that the outliers do not affect the results of our method and then apply Factor Analysis using the extraction method of Principal Components. We also mention in detail all steps until reaching our final conclusions.

Submitted by Automa??o e Estat?stica (sst@bczm.ufrn.br) on 2017-11-22T20:38:08Z No. of bitstreams: 1 ClauberGomesBezerra_TESE.pdf: 7864722 bytes, checksum: 17c21362443f4d25511a0a211d52b805 (MD5) / Approved for entry into archive by Arlan Eloi Leite Silva (eloihistoriador@yahoo.com.br) on 2017-11-23T23:24:44Z (GMT) No. of bitstreams: 1 ClauberGomesBezerra_TESE.pdf: 7864722 bytes, checksum: 17c21362443f4d25511a0a211d52b805 (MD5) / Made available in DSpace on 2017-11-23T23:24:44Z (GMT). No. of bitstreams: 1 ClauberGomesBezerra_TESE.pdf: 7864722 bytes, checksum: 17c21362443f4d25511a0a211d52b805 (MD5) Previous issue date: 2017-05-24 / Nesta tese apresentamos uma nova abordagem para realizar o agrupamento e a classifica??o de um conjunto de dados de forma n?o supervisionada. A abordagem proposta utiliza os conceitos de tipicidade e excentricidade usados pelo algoritmo TEDA na detec??o de outliers. Para realizar o agrupamento e a classifica??o ? proposto um algoritmo estat?stico chamado Auto-Cloud. As amostras analisadas pelo Auto-Cloud s?o agrupadas em unidades chamadas de data clouds, que s?o estruturas que n?o possuem formato ou limites definidos. O Auto-Cloud permite que cada amostra analisada possa pertencer simultaneamente a v?rias data clouds. O Auto-Cloud ? um algoritmo aut?nomo e evolutivo, que n?o necessita de treinamento ou qualquer conhecimento pr?vios sobre o conjunto de dados analisado. Ele permite a cria??o e a fus?o das data clouds de forma aut?noma, ? medida que as amostras s?o lidas, sem qualquer interven??o humana. As caracter?sticas do algoritmo fazem com que ele seja indicado para o agrupamento e classifica??o de streams de dados e para aplica??es que requerem respostas em tempo-real. O Auto- Cloud tamb?m ? um algoritmo recursivo, o que o torna r?pido e exige pouca quantidade de mem?ria. J? no processo de classifica??o dos dados, o Auto-Cloud trabalha como um classificador fuzzy, calculando o grau de pertin?ncia entre a amostra analisada e cada data cloud criada no processo de agrupamento. A classe a que pertence cada amostra ? determinada pela data cloud com maior grau de pertin?ncia com rela??o a amostra. Para validar o m?todo proposto, aplicamos o mesmo em v?rios conjuntos de dados existentes na literatura sobre o assunto. Al?m disso, o m?todo tamb?m foi validado numa aplica??o de detec??o e classifica??o de falhas em processos industriais, onde foram utilizados dados reais, obtidos de uma planta industrial. / In this thesis we propose a new approach to unsupervised data clustering and classification. The proposed approach is based on typicality and eccentricity concepts. This concepts are used by recently introduced TEDA algorithm for outlier detection. To perform data clustering and classification, it is proposed a new statistical algorithm, called Auto-Cloud. The data samples analyzed by Auto-Cloud are grouped in the form of unities called data clouds, which are structures without pre-defined shape or boundaries. Auto-Cloud allows each data sample to belong to multiple data clouds simultaneously. Auto-Cloud is an autonomous and evolving algorithm, which does not requires previous training or any prior knowledge about the data set. Auto-Cloud is able to create and merge data clouds autonomously, as data samples are obtained, without any human interference. The algorithm is suitable for data clustering and classification of online data streams and application that require real-time response. Auto-Cloud is also recursive, which makes it fast and with little computational effort. The data classification process works like a fuzzy classifier using the degree of membership between the analyzed data sample to each data cloud created in clustering process. The class to which each data sample belongs is determined by the cloud with the highest activation with respect to that sample. To validate the proposed method, we apply it to several existing datasets for data clustering and classification. Moreover, the method was also used in a fault detection in industrial processes application. In this case, we use real data obtained from a real world industrial plant.

Após o ajustamento de observações pelo método dos mínimos quadrados (MMQ) ter sido realizado, é possível a detecção e a identificação de erros não aleatórios nas observações, por meio de testes estatísticos. A teoria da confiabilidade faz uso de medidas adequadas para quantificar o menor erro detectável em uma observação, e a sua influência sobre os parâmetros ajustados, quando não detectado. A teoria de confiabilidade convencional foi desenvolvida para os procedimentos de teste convencionais, como o data snooping, que pressupõem que apenas uma observação está contaminada por erros grosseiros por vez. Recentemente foram desenvolvidas medidas de confiabilidade generalizadas, relativas a testes estatísticos que pressupõem a existência, simultânea, de múltiplas observações com erros (outliers). Outras abordagens para o controle de qualidade do ajustamento, alternativas a estes testes estatísticos, também foram propostas recentemente, como por exemplo, o método QUAD (Quasi-Accurate Detection of outliers method). Esta pesquisa tem por objetivo fazer um estudo sobre o controle de qualidade do ajustamento de observações geodésicas, por meio de experimentos em uma rede GPS (Global Positioning System), utilizando tanto os métodos convencionais quanto o atual estado da arte. Desta forma, foram feitos estudos comparativos entre medidas de confiabilidade convencionais e medidas de confiabilidade generalizadas para dois outliers simultâneos, bem como estudos comparativos entre o procedimento data snooping e testes estatísticos para a identificação de múltiplos outliers. Também se investigou como a questão das variâncias e covariâncias das observações, bem como a geometria/configuração da rede GPS em estudo, podem influenciar nas medidas de confiabilidade, tanto na abordagem convencional, quanto na abordagem generalizada. Por fim, foi feito um estudo comparativo entre o método QUAD e os testes estatísticos para a identificação de erros. / After the adjustment of observations has been carried out by Least Squares Method (LSM), it is possible to detect and identify non-random errors in the observations using statistical tests. The reliability theory makes use of appropriate measures to quantify the minimal detectable bias (error) in an observation, and its influence on the adjusted parameters, if not detected. The conventional reliability theory has been developed for conventional testing procedures such as data snooping, which assumes that only one observation is contaminated by errors at a time. Recently, generalized measures of reliability has been developed, relating to statistical tests that assumes the existence, simultaneous, of multiple observations with errors (outliers). Other approaches to the quality control of the adjustment, alternatives to these statistical tests, were also proposed recently, such as the QUAD method (Quasi-Accurate Detection of outliers method). The goal of this research is to make a study about the quality control of the adjustment of geodetic observations, by means of experiments in a GPS (Global Positioning System) network, using both conventional methods and the current state of the art. In this way, comparisons were made between conventional reliability measures and generalized measures of reliability for two outliers, as well as comparisons between the data snooping procedure and statistical tests to identify multiple outliers. It was also investigated how the variances and covariances of the observations, as well as the geometry/configuration of the GPS network in study, can influence the measures of reliability, both in the conventional approach and in the generalized approach. Finally, a comparison was made between the QUAD method and the statistical tests to identify outliers (errors).

Submitted by Elizabete Silva (elizabete.silva@ufes.br) on 2015-11-23T18:55:03Z No. of bitstreams: 2 license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) ANALISE FATORIAL EM SERIES TEMPORAIS.pdf: 2194722 bytes, checksum: 443c7c57567200ac6397234fc6b5687f (MD5) / Approved for entry into archive by Morgana Andrade (morgana.andrade@ufes.br) on 2016-01-05T10:06:44Z (GMT) No. of bitstreams: 2 license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) ANALISE FATORIAL EM SERIES TEMPORAIS.pdf: 2194722 bytes, checksum: 443c7c57567200ac6397234fc6b5687f (MD5) / Made available in DSpace on 2016-01-05T10:06:44Z (GMT). No. of bitstreams: 2 license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) ANALISE FATORIAL EM SERIES TEMPORAIS.pdf: 2194722 bytes, checksum: 443c7c57567200ac6397234fc6b5687f (MD5) Previous issue date: 2015 / CAPES / Os estudos de polui c~ao atmosf erica geralmente envolvem medi c~oes e an alises de dados de concentra c~oes de poluentes, como e o caso do MP10 (material particulado), de SO2 (di oxido de enxofre) e de outros poluentes. Estes dados normalmente possuem caracter sticas importantes como autocorrela c~ao, longa depend^encia, sazonalidade e observa c~oes at picas, que necessitam de ferramentas de an alise de s eries temporais multivariadas para avaliar o seu comportamento na atmosfera. Neste contexto, propomos um estimador fracion ario robusto da matriz de autocovari^ancia robusta de longa depend^encia e frequ^encia sazonal, para o modelo SARFIMA. O interesse pr atico em polui c~ao do ar e avaliar o comportamento das s eries de concentra c~oes de SO2 e fazer as previs~oes, mais acuradas, deste poluente. As previs~oes, do modelo SARFIMA estimado, s~ao comparadas as previs~oes do modelo SARMA, atrav es do erro quadr atico m edio. Existe outra di culdade na investiga c~ao dos poluentes atmosf ericos, por modelos de s eries temporais: os dados de SO2, de MP10 e de outros poluentes possuem alta dimensionalidade. Este fato di culta o tratamento dos dados atrav es de modelos vetoriais autorregressivos, pelo excessivo n umero de par^ametros estimados. Na literatura, a abordagem do problema para s eries temporais de grandes dimens~oes e feita atrav es da redu c~ao da dimensionalidade dos dados, utilizando, principalmente, o modelo fatorial e o m etodo de componentes principais. Por em, as caracter sticas de longa depend^encia e de observa c~oes at picas das s eries de polui c~ao atmosf erica, normalmente, n~ao s~ao envolvidas na teoria de an alise fatorial. Neste contexto, propomos aqui uma contribui c~ao te orica para o modelo fatorial de s eries temporais de grandes dimens~oes, envolvendo longa depend^encia e robustez na estima c~ao dos fatores. O modelo sugerido e aplicado em s eries de MP10 da rede de monitoramento da qualidade do ar da Grande Vit oria - ES. / Studies about air pollution typically involve measurements and analysis of pollutants, such as PM10 (particulate matter), SO2 (sulfur dioxide) and others. These data typically have important features like serial correlation, long dependency, seasonality and occurence of atypical observations, and many others, which may be analyzed by means of multivariate time series. In this context, a robust estimator of fractional robust autocovariance matrix of long dependence and seasonal frequency for SARFIMA model is proposed. The model is compared to SARMA model and is applied to SO2 concentrations. In addition of the mentioned features the data present high dimensionality in relation to sample size and number of variables. This fact complicates the analisys of the data using vector time series models. In the literature, the approach to mitigate this problem for high dimensional time series is to reduce the dimensionality using the factor analysis and principal component analysis. However, the long dependence characteristics and atypical observations, very common in air pollution series, is not considered by the standard factor analysis method. In this context, the standard factor model is extended to consider time series data presenting long dependence and outliers. The proposed method is applied to PM10 series of air quality monitoring network of the Greater Vit oria Region - ES.

Parmi tous les leviers à disposition des entreprises, la prise en compte de la perception par les clients est aujourd'hui centrale, dès la conception des produits. En effet, le consommateur est aujourd'hui mieux informé et attentif à ce qu'il perçoit de la qualité d'un produit et cette perception lui permet d'établir une valeur d'estime de la qualité esthétique des produits, mais aussi de ses fonctionnalités techniques. La méthodologie de l'analyse de la qualité d'aspect des surfaces est donc un enjeu essentiel pour l'industrie. Deux approches de la fonctionnalité des surfaces sont proposées afin de formaliser la méthodologie de détection, et d'apporter aux experts des critères objectifs d'évaluation des anomalies. La première approche proposée est basée sur la métrologie des surfaces. Elle consiste à analyser les topographies mesurées pour lier la fonction aspect aux caractéristiques géométriques extraites. Une approche multi-échelle basée sur la Décomposition Modale Discrète est mise en oeuvre afin de séparer efficacement les différents ordres de variations géométriques d'une surface, et ainsi d'isoler les anomalies d'aspect. D'autre part, cette méthode permet la mise en oeuvre du calcul des courbures sur une surface de façon simplifiée et robuste. On montre que cet attribut géométrique apporte une information supplémentaire et pertinente en lien avec la fonction aspect. Enfin, ces travaux ont mis en évidence l'importance de la qualité des données sources pour analyser l'aspect, et particulièrement deux difficultés d'ordre métrologiques, liées à la présence de points aberrants (hautes fréquences) et de variations géométriques non intrinsèques aux surfaces, générées par le moyen de mesure (basses fréquences). Une méthode innovante d'identification des points aberrants dédiée à la métrologie des surfaces et basée sur une approche statistique multi-échelle est proposée. La problématique des variations géométriques liées aux tables de positionnement du moyen de mesure est traitée au moyen de la Décomposition Modale, et un protocole pour corriger ces variations est présenté. La seconde approche, plus globale, est basée sur l'interaction entre les surfaces et l'environnement lumineux. L'objet de cette approche de l'analyse de l'aspect est d'apporter une aide aux experts pour mieux détecter les anomalies. Les travaux présentés sont basés sur la technique Polynomial Texture Mappings et consistent à modéliser la réflectance en chaque point des surfaces afin de simuler le rendu visuel sous un éclairage quelconque, à la manière de ce que font les opérateurs en analyse sensorielle pour faciliter la détection. Un dispositif d'aide à l'inspection des surfaces basé sur ce principe est présenté. Enfin, une approche industrielle est proposée afin de montrer comment ces 2 axes de recherche peuvent être complémentaires dans le cadre d'une méthodologie globale, industrielle, de l'analyse de la qualité d'aspect de surfaces. / Accounting for customers' perception of manufactured goods has become a major challenge for the industry. This process is to be established from early design to retail. Customers are nowadays more aware and detail oriented about perceived quality of products. This allows one to set not only an estimated price but also the expected quality of the product. Surface appearance analysis has therefore become a key industrial issue. Two approaches are proposed here to formalize the detection methodology and provide objective criteria for experts to evaluate surface anomalies. The first proposed approach is based on surface metrology. It consists in analyzing the measured topologies in order to bind aspect to geometric characteristics. A multi-scale procedure based on Discrete Modal Decomposition is implemented and allows an effective separation of geometric variations. Accordingly, appearance anomalies can be isolated from other geometrical features. This method enables the calculation of surface curvatures in a simplified and robust manner. It is shown that such geometric information is relevant and bound to visual aspect. The presented work also emphasizes the influence of raw data in aspect analysis. Two main metrological difficulties are investigated: the presence of outliers (High frequencies) and the presence of non surface-related geometric defects, generated by the measuring device (Low frequencies). An innovative method for identifying outliers in surface metrology is presented. It is based on a multi-scale statistical approach. Finally, the issue of geometrical variation due to positioning tables is also addressed. A calibration protocol based on DMD that intends to correct this phenomenon is proposed. The second proposed approach, more global, is based on the interaction of a surface with its light environment. It aims at providing experts with assistance, specifically during the anomaly detection phase. The presented work uses Polynomial Texture Mapping. This technique consists of calculating the reflectance at each point of the surface and simulating its appearance while the lighting angles vary. A surface Inspection Support Device based on this principle is presented and detailed. Finally, an industrial study is proposed that shows how these two academic approaches can be combined within a global industrial methodology dedicated to surface appearance quality.

L'estimation précise du mouvement 3D d'une caméra relativement à une scène rigideest essentielle pour tous les systèmes de navigation visuels. Aujourd'hui différentstypes de capteurs sont adoptés pour se localiser et naviguer dans des environnementsinconnus : GPS, capteurs de distance, caméras, capteurs magnétiques, centralesinertielles (IMU, Inertial Measurement Unit). Afin d'avoir une estimation robuste, lesmesures de plusieurs capteurs sont fusionnées. Même si le progrès technologiquepermet d'avoir des capteurs de plus en plus précis, et si la communauté de robotiquemobile développe algorithmes de navigation de plus en plus performantes, il y aencore des défis ouverts. De plus, l'intérêt croissant des la communauté de robotiquepour les micro robots et essaim de robots pousse vers l'emploi de capteurs à bas poids,bas coût et vers l'étude d'algorithmes à faible complexité. Dans ce contexte, capteursinertiels et caméras monoculaires, grâce à leurs caractéristiques complémentaires,faible poids, bas coût et utilisation généralisée, représentent une combinaison decapteur intéressante.Cette thèse présente une contribution dans le cadre de la navigation inertielle assistéepar vision et aborde les problèmes de fusion de données et estimation de la pose, envisant des algorithmes à faible complexité appliqués à des micro-véhicules aériens.Pour ce qui concerne l'association de données, une nouvelle méthode pour estimer lemouvement relatif entre deux vues de caméra consécutifs est proposée.Celle-ci ne nécessite l'observation que d'un seul point caractéristique de la scène et laconnaissance des vitesses angulaires fournies par la centrale inertielle, sousl'hypothèse que le mouvement de la caméra appartient localement à un planperpendiculaire à la direction de la gravité. Deux algorithmes très efficaces pouréliminer les fausses associations de données (outliers) sont proposés sur la base decette hypothèse de mouvement.Afin de généraliser l'approche pour des mouvements à six degrés de liberté, deuxpoints caracteristiques et les données gyroscopiques correspondantes sont nécessaires.Dans ce cas, deux algorithmes sont proposés pour éliminer les outliers.Nous montrons que dans le cas d'une caméra monoculaire montée sur un quadrotor,les informations de mouvement fournies par l'IMU peuvent être utilisées pouréliminer de mauvaises estimations.Pour ce qui concerne le problème d'estimation de la pose, cette thèse fournit unesolution analytique pour exprimer la pose du système à partir de l'observation de troispoints caractéristiques naturels dans une seule image, une fois que le roulis et letangage sont obtenus par les données inertielles sous l'hypothèse de terrain plan.Afin d'aborder le problème d'estimation de la pose dans des environnements sombresou manquant de points caractéristiques, un système équipé d'une caméra monoculaire,d'une centrale inertielle et d'un pointeur laser est considéré. Grace à une analysed'observabilité il est démontré que les grandeurs physiques qui peuvent êtredéterminées par l'exploitation des mesures fourni par ce systeme de capteurs pendantun court intervalle de temps sont : la distance entre le système et la surface plane ;la composante de la vitesse du système qui est orthogonale à la surface ; l'orientationrelative du système par rapport à la surface et l'orientation de la surface par rapport àla gravité. Une méthode récursive simple a été proposée pour l'estimation de toutesces quantités observables.Toutes les contributions de cette thèse sont validées par des expérimentations à l'aidedes données réelles et simulées. Grace à leur faible complexité de calcul, lesalgorithmes proposés sont très appropriés pour la mise en oeuvre en temps réel surdes systèmes ayant des ressources de calcul limitées. La suite de capteur considéréeest monté sur un quadrotor, mais les contributions de cette thèse peuvent êtreappliquées à n'importe quel appareil mobile. / Accurate egomotion estimation is of utmost importance for any navigation system.Nowadays di_erent sensors are adopted to localize and navigate in unknownenvironments such as GPS, range sensors, cameras, magnetic field sensors, inertialsensors (IMU). In order to have a robust egomotion estimation, the information ofmultiple sensors is fused. Although the improvements of technology in providingmore accurate sensors, and the efforts of the mobile robotics community in thedevelopment of more performant navigation algorithms, there are still openchallenges. Furthermore, the growing interest of the robotics community in microrobots and swarm of robots pushes towards the employment of low weight, low costsensors and low computational complexity algorithms. In this context inertial sensorsand monocular cameras, thanks to their complementary characteristics, low weight,low cost and widespread use, represent an interesting sensor suite.This dissertation represents a contribution in the framework of vision-aided inertialnavigation and tackles the problems of data association and pose estimation aimingfor low computational complexity algorithms applied to MAVs.For what concerns the data association, a novel method to estimate the relative motionbetween two consecutive camera views is proposed. It only requires the observationof a single feature in the scene and the knowledge of the angular rates from an IMU,under the assumption that the local camera motion lies in a plane perpendicular to thegravity vector. Two very efficient algorithms to remove the outliers of the featurematchingprocess are provided under the abovementioned motion assumption. Inorder to generalize the approach to a 6DoF motion, two feature correspondences andgyroscopic data from IMU measurements are necessary. In this case, two algorithmsare provided to remove wrong data associations in the feature-matching process. Inthe case of a monocular camera mounted on a quadrotor vehicle, motion priors fromIMU are used to discard wrong estimations.For what concerns the pose estimation problem, this thesis provides a closed formsolution which gives the system pose from three natural features observed in a singlecamera image, once the roll and the pitch angles are obtained by the inertialmeasurements under the planar ground assumption.In order to tackle the pose estimation problem in dark or featureless environments, asystem equipped with a monocular camera, inertial sensors and a laser pointer isconsidered. The system moves in the surrounding of a planar surface and the laserpointer produces a laser spot on the abovementioned surface. The laser spot isobserved by the monocular camera and represents the only point feature considered.Through an observability analysis it is demonstrated that the physical quantities whichcan be determined by exploiting the measurements provided by the aforementionedsensor suite during a short time interval are: the distance of the system from the planarsurface; the component of the system speed that is orthogonal to the planar surface;the relative orientation of the system with respect to the planar surface; the orientationof the planar surface with respect to the gravity. A simple recursive method toperform the estimation of all the aforementioned observable quantities is provided.All the contributions of this thesis are validated through experimental results usingboth simulated and real data. Thanks to their low computational complexity, theproposed algorithms are very suitable for real time implementation on systems withlimited on-board computation resources. The considered sensor suite is mounted on aquadrotor vehicle but the contributions of this dissertations can be applied to anymobile device.

Submitted by Automa??o e Estat?stica (sst@bczm.ufrn.br) on 2016-06-10T00:04:44Z No. of bitstreams: 1 LeandroLuttianeDaSilvaLinhares_TESE.pdf: 2400561 bytes, checksum: 3693662adcc0c23b5063f51b23d9b6c5 (MD5) / Approved for entry into archive by Arlan Eloi Leite Silva (eloihistoriador@yahoo.com.br) on 2016-06-10T22:50:22Z (GMT) No. of bitstreams: 1 LeandroLuttianeDaSilvaLinhares_TESE.pdf: 2400561 bytes, checksum: 3693662adcc0c23b5063f51b23d9b6c5 (MD5) / Made available in DSpace on 2016-06-10T22:50:22Z (GMT). No. of bitstreams: 1 LeandroLuttianeDaSilvaLinhares_TESE.pdf: 2400561 bytes, checksum: 3693662adcc0c23b5063f51b23d9b6c5 (MD5) Previous issue date: 2015-09-03 / Coordena??o de Aperfei?oamento de Pessoal de N?vel Superior (CAPES) / O grande interesse pela identifica??o n?o linear de sistemas din?micos deve-se, principalmente, ao fato de que uma grande quantidade dos sistemas reais s?o complexos e precisam ter suas n?o linearidades consideradas para que seus modelos possam ser utilizados com sucesso em aplica??es, por exemplo, de controle, predi??o, infer?ncia, entre outros. O presente trabalho analisa a aplica??o das redes Fuzzy Wavelet Neural Network (FWNN) na identifica??o de sistemas n?o lineares sujeitos a ru?dos e outliers. Esses elementos, geralmente, influenciam no procedimento de identifica??o, ocasionando interpreta??es err?neas em rela??o ao comportamento din?mico do sistema. A FWNN combina, em uma ?nica estrutura, a capacidade de tratar incertezas da l?gica fuzzy, as caracter?sticas de multirresolu??o da teoria wavelet e as habilidades de aprendizado e generaliza??o das redes neurais artificiais. Normalmente, o aprendizado dessas redes ? realizado por algum m?todo baseado em gradiente, tendo o erro m?dio quadr?tico como fun??o de custo. Este trabalho avalia a substitui??o dessa tradicional fun??o por uma medida de similaridade da Teoria da Informa??o, denominada correntropia. Esta medida de similaridade permite que momentos estat?sticos de ordem superior possam ser considerados durante o processo de treinamento. Por esta raz?o, ela se torna mais apropriada para distribui??es de erro n?o gaussianas e faz com que o treinamento apresente menor sensibilidade ? presen?a de outliers. Para avaliar esta substitui??o, modelos FWNN s?o obtidos na identifica??o de dois estudos de caso: um sistema real n?o linear, consistindo em um tanque de m?ltiplas se??es, e um sistema simulado baseado em um modelo biomec?nico da articula??o do joelho humano. Os resultados obtidos demonstram que a utiliza??o da correntropia, como fun??o custo no algoritmo de retropropaga??o do erro, torna o procedimento de identifica??o utilizando redes FWNN mais robusto aos outliers. Entretanto, isto somente pode ser alcan?ado a partir do ajuste adequado da largura do kernel gaussiano da correntropia. / The great interest in nonlinear system identification is mainly due to the fact that a large amount of real systems are complex and need to have their nonlinearities considered so that their models can be successfully used in applications of control, prediction, inference, among others. This work evaluates the application of Fuzzy Wavelet Neural Networks (FWNN) to identify nonlinear dynamical systems subjected to noise and outliers. Generally, these elements cause negative effects on the identification procedure, resulting in erroneous interpretations regarding the dynamical behavior of the system. The FWNN combines in a single structure the ability to deal with uncertainties of fuzzy logic, the multiresolution characteristics of wavelet theory and learning and generalization abilities of the artificial neural networks. Usually, the learning procedure of these neural networks is realized by a gradient based method, which uses the mean squared error as its cost function. This work proposes the replacement of this traditional function by an Information Theoretic Learning similarity measure, called correntropy. With the use of this similarity measure, higher order statistics can be considered during the FWNN training process. For this reason, this measure is more suitable for non-Gaussian error distributions and makes the training less sensitive to the presence of outliers. In order to evaluate this replacement, FWNN models are obtained in two identification case studies: a real nonlinear system, consisting of a multisection tank, and a simulated system based on a model of the human knee joint. The results demonstrate that the application of correntropy as the error backpropagation algorithm cost function makes the identification procedure using FWNN models more robust to outliers. However, this is only achieved if the gaussian kernel width of correntropy is properly adjusted.

Após o ajustamento de observações pelo método dos mínimos quadrados (MMQ) ter sido realizado, é possível a detecção e a identificação de erros não aleatórios nas observações, por meio de testes estatísticos. A teoria da confiabilidade faz uso de medidas adequadas para quantificar o menor erro detectável em uma observação, e a sua influência sobre os parâmetros ajustados, quando não detectado. A teoria de confiabilidade convencional foi desenvolvida para os procedimentos de teste convencionais, como o data snooping, que pressupõem que apenas uma observação está contaminada por erros grosseiros por vez. Recentemente foram desenvolvidas medidas de confiabilidade generalizadas, relativas a testes estatísticos que pressupõem a existência, simultânea, de múltiplas observações com erros (outliers). Outras abordagens para o controle de qualidade do ajustamento, alternativas a estes testes estatísticos, também foram propostas recentemente, como por exemplo, o método QUAD (Quasi-Accurate Detection of outliers method). Esta pesquisa tem por objetivo fazer um estudo sobre o controle de qualidade do ajustamento de observações geodésicas, por meio de experimentos em uma rede GPS (Global Positioning System), utilizando tanto os métodos convencionais quanto o atual estado da arte. Desta forma, foram feitos estudos comparativos entre medidas de confiabilidade convencionais e medidas de confiabilidade generalizadas para dois outliers simultâneos, bem como estudos comparativos entre o procedimento data snooping e testes estatísticos para a identificação de múltiplos outliers. Também se investigou como a questão das variâncias e covariâncias das observações, bem como a geometria/configuração da rede GPS em estudo, podem influenciar nas medidas de confiabilidade, tanto na abordagem convencional, quanto na abordagem generalizada. Por fim, foi feito um estudo comparativo entre o método QUAD e os testes estatísticos para a identificação de erros. / After the adjustment of observations has been carried out by Least Squares Method (LSM), it is possible to detect and identify non-random errors in the observations using statistical tests. The reliability theory makes use of appropriate measures to quantify the minimal detectable bias (error) in an observation, and its influence on the adjusted parameters, if not detected. The conventional reliability theory has been developed for conventional testing procedures such as data snooping, which assumes that only one observation is contaminated by errors at a time. Recently, generalized measures of reliability has been developed, relating to statistical tests that assumes the existence, simultaneous, of multiple observations with errors (outliers). Other approaches to the quality control of the adjustment, alternatives to these statistical tests, were also proposed recently, such as the QUAD method (Quasi-Accurate Detection of outliers method). The goal of this research is to make a study about the quality control of the adjustment of geodetic observations, by means of experiments in a GPS (Global Positioning System) network, using both conventional methods and the current state of the art. In this way, comparisons were made between conventional reliability measures and generalized measures of reliability for two outliers, as well as comparisons between the data snooping procedure and statistical tests to identify multiple outliers. It was also investigated how the variances and covariances of the observations, as well as the geometry/configuration of the GPS network in study, can influence the measures of reliability, both in the conventional approach and in the generalized approach. Finally, a comparison was made between the QUAD method and the statistical tests to identify outliers (errors).