This thesis is devoted to the problems of defining and developing the basic building blocks of an automated surveillance system. As its initial step, a background-modeling algorithm is described for segmenting moving objects from the background, which is capable of adapting to dynamic scene conditions, as well as determining shadows of the moving objects. After obtaining binary silhouettes for targets, object association between consecutive frames is achieved by a hypothesis-based tracking method. Both of these tasks provide basic information for higher-level processing, such as activity analysis and object identification. In order to recognize the nature of an event occurring in a scene, hidden Markov models (HMM) are utilized. For this aim, object trajectories, which are obtained through a successful track, are written as a sequence of flow vectors that capture the details of instantaneous velocity and location information. HMMs are trained with sequences obtained from usual motion patterns and abnormality is detected by measuring the distance to these models. Finally, MPEG-7 visual descriptors are utilized in a regional manner for object identification. Color structure and homogeneous texture parameters of the independently moving objects are extracted and classifiers, such as Support Vector Machine (SVM) and Bayesian plug-in (Mahalanobis distance), are utilized to test the performance of the proposed person identification mechanism. The simulation results with all the above building blocks give promising results, indicating the possibility of constructing a fully automated surveillance system for the future.

In this study, a new discriminative learning framework, called soft margin estimation (SME), is proposed for estimating the parameters of continuous density hidden Markov models (HMMs). The proposed method makes direct use of the successful ideas of margin in support vector machines to improve generalization capability and decision feedback learning in discriminative training to enhance model separation in classifier design. SME directly maximizes the separation of competing models to enhance the testing samples to approach a correct decision if the deviation from training samples is within a safe margin. Frame and utterance selections are integrated into a unified framework to select the training utterances and frames critical for discriminating competing models. SME offers a flexible and rigorous framework to facilitate the incorporation of new margin-based optimization criteria into HMMs training. The choice of various loss functions is illustrated and different kinds of separation measures are defined under a unified SME framework. SME is also shown to be able to jointly optimize feature extraction and HMMs. Both the generalized probabilistic descent algorithm and the Extended Baum-Welch algorithm are applied to solve SME. SME has demonstrated its great advantage over other discriminative training methods in several speech recognition tasks. Tested on the TIDIGITS digit recognition task, the proposed SME approach achieves a string accuracy of 99.61%, the best result ever reported in literature. On the 5k-word Wall Street Journal task, SME reduced the word error rate (WER) from 5.06% of MLE models to 3.81%, with relative 25% WER reduction. This is the first attempt to show the effectiveness of margin-based acoustic modeling for large vocabulary continuous speech recognition in a HMMs framework. The generalization of SME was also well demonstrated on the Aurora 2 robust speech recognition task, with around 30% relative WER reduction from the clean-trained baseline.

A piece of popular music is usually defined as a combination of vocal melody and instrumental accompaniment. People often start with the melody part when they are trying to compose or reproduce a piece of popular music. However, creating appropriate instrumental accompaniment part for a melody line can be a difficult task for non-musicians. Automation of accompaniment generation for vocal melodies thus can be very useful for those who are interested in singing for fun. Therefore, a computer software system which is capable of generating harmonic accompaniment for a given vocal melody input has been presented in this thesis. This automatic accompaniment system uses a Hidden Markov Model to assign chord to a given part of melody based on the knowledge learnt from a bank of vocal tracks of popular music. Comparing with other similar systems, our system features a high resolution key estimation algorithm which is helpful to adjust the generated accompaniment to the input vocal. Moreover, we designed a structure analysis subsystem to extract the repetition and structure boundaries from the melody. These boundaries are passed to the chord assignment and style player subsystems in order to generate more dynamic and organized accompaniment. Finally, prototype applications are discussed and the entire system is evaluated.

Les Interfaces Cerveau-Machine (ICM) sont des systèmes qui permettent à des patients souffrant d'un handicap moteur sévère d'utiliser leur activité cérébrale pour contrôler des effecteurs, par exemple des prothèses des membres supérieurs dans le cas d'ICM motrices. Les intentions de mouvement de l'utilisateur sont estimées en appliquant un décodeur sur des caractéristiques extraites de son activité cérébrale. Des challenges spécifiques au déploiement clinique d'ICMs motrices ont été considérés, à savoir le contrôle mono-membre ou séquentiel multi-membre asynchrone et précis. Un décodeur, le Markov Switching Linear Model (MSLM), a été développé pour limiter les activations erronées de l'ICM, empêcher des mouvements parallèles des effecteurs et décoder avec précision des mouvements complexes. Le MSLM associe des modèles linéaires à différents états possibles, e.g. le contrôle d'un membre spécifique ou une phase de mouvement particulière. Le MSLM réalise une détection d'état dynamique, et les probabilités des états sont utilisées pour pondérer les modèles linéaires.La performance du décodeur MSLM a été évaluée pour la reconstruction asynchrone de trajectoires de poignet et de doigts à partir de signaux electrocorticographiques. Il a permis de limiter les activations erronées du système et d'améliorer la précision du décodage du signal cérébral. / Brain-Computer Interfaces (BCI) are systems that allow severely motor-impaired patients to use their brain activity to control external devices, for example upper-limb prostheses in the case of motor BCIs. The user's intentions are estimated by applying a decoder on neural features extracted from the user's brain activity. Signal processing challenges specific to the clinical deployment of motor BCI systems are addressed in the present doctoral thesis, namely asynchronous mono-limb or sequential multi-limb decoding and accurate decoding during active control states. A switching decoder, namely a Markov Switching Linear Model (MSLM), has been developed to limit spurious system activations, to prevent parallel limb movements and to accurately decode complex movements.The MSLM associates linear models with different possible control states, e.g. activation of a specific limb, specific movement phases. Dynamic state detection is performed by the MSLM, and the probability of each state is used to weight the linear models. The performance of the MSLM decoder was assessed for asynchronous wrist and multi-finger trajectory reconstruction from electrocorticographic signals. It was found to outperform previously reported decoders for the limitation of spurious activations during no-control periods and permitted to improve decoding accuracy during active periods.

[pt] A ocorrência de erros em surto é observada principalmente em canais sem fio. Para a análise e melhor entendimento deste tipo de erro, a fim de se melhorar os projetos de sistemas de comunicações digitais, uma modelagem mais precisa, de canais com esta característica, torna-se necessária. Uma diversidade de métodos de estimação de parâmetros tem sido estudada, principalmente aquelas baseadas nos Modelos Escondidos de Markov (HMM do ingês). Em geral cada método é focado em um sistema de comunicações específico, sobre uma camada específica. Neste trabalho é proposto um novo método baseado em um HMM com uma estrutura particular, que permite a dedução de expressões analíticas para todas as estatísticas de interesse. A estrutura do modelo proposto permite a geração de eventos que ocorrem numa sequência binária de dados sujeita a surtos de erro, de acordo com a definição de surtos de erro do CCITT. O modelo proposto possui um número fixo de apenas sete parâmetros, mas o seu número de estados cresce com um de seus parâmetros, que aumenta a precisão, mas não a complexidade. Este trabalho adotou técnicas de otimização, associadas aos métodos de Máxima Verossimilhança e Particle Swarm Optimization (PSO) a fim de realizar a estimação dos parâmetros do modelo proposto. Os resultados demonstram que o modelo proposto permite a caracterização precisa de canais com memória de diversas origens. / [en] The occurrence of error busts is mainly observed in wireless channels. For analysis and a better understanding of such errors, in order to improve the design of communication systems, an accurate modeling of channels with this characteristic is necessary. A lot of parameter estimation methods have been studied, mainly the ones based on Hidden Markov Models (HMM). In general each method is focused in a specific communication system, on a specific layer. In this work it is proposed a new method based on a HMM with particular structure that allows the deduction of analytical expressions for all statistics of interest. The structure of the proposed model permits the generation of events that occur in a binary data sequence subject to bursts of error concerning CCITT error burst definition. The proposed model has a fixed number of only seven parameters but its number of states increase with one of those parameters that increase the accuracy but not the complexity. This work adopted techniques of optimization associated to Maximum Likelihood (ML) and Particle Swarm Optimization (PSO) to perform the parameter estimation to the proposed model. The results show that the proposed model achieves accurate characterization of channels with memory from many different sources.

Made available in DSpace on 2014-06-11T19:29:40Z (GMT). No. of bitstreams: 0 Previous issue date: 2011-08-05Bitstream added on 2014-06-13T19:38:57Z : No. of bitstreams: 1 kami_gjc_me_sjrp.pdf: 1342570 bytes, checksum: 240c6d6b92fda1861dfbed94c9213a10 (MD5) / As técnicas para identificação biométrica têm evoluído cada vez mais devido à necessidade que os seres humanos têm de identificar as pessoas em tempo real e de forma precisa para permitir o acesso a determinados recursos, como por exemplo, as aplicações e serviços WEB. O reconhecimento facial é uma técnica biométrica que apresenta várias vantagens em relação às demais, tais como: uso de equipamentos simples e baratos para a obtenção das amostras e a possibilidade de se realizar o reconhecimento em sigilo e à distância. O reconhecimento de faces a partir de vídeo é uma tendência recente na área de Biometria. Esta dissertação tem por objetivo principal comparar diferentes técnicas de reconhecimento facial a partir de vídeo para determinar as que apresentam um melhor compromisso entre tempo de processamento e precisão. Outro objetivo é a incorporação dessas melhores técnicas no sistema de autenticação biométrica em ambientes de E-Learning, proposto em um trabalho anterior. Foi comparado o classificador vizinho mais próximo usando as medidas de distância Euclidiana e Mahalanobis com os seguintes classificadores: Redes Neurais MLP e SOM, K Vizinhos mais Próximos, Classificador Bayesiano, Máquinas de Vetores de Suporte (SVM) e Floresta de Caminhos Ótimos (OPF). Também foi avaliada a técnica de Modelos Ocultos de Markov (HMM). Nos experimentos realizados com a base Recogna Video Database, criada especialmente para uso neste trabalho, e Honda/UCSD Video Database, os classificadores apresentaram os melhores resultados em termos de precisão, com destaque para o classificador SVM da biblioteca SVM Torch. A técnica HMM, que incorpora informações temporais, apresentou resultados melhores do que as funções de distância, em termos de precisão, mas inferiores aos classificadores / The biometric identification techniques have evolved increasingly due to the need that humans have to identify people in real time to allow access to certain resources, such as applications and Web services. Facial recognition is a biometric technique that has several advantages over others. Some of these advantages are the use of simple and cheap equipment to obtain the samples and the ability to perform the recognition in covert mode. The face recognition from video is a recent approach in the area of Biometrics. The work in this dissertation aims at comparing different techniques for face recognition from video in order to find the best rates on processing time and accuracy. Another goal is the incorporation of these techniques in the biometric authentication system for E-Learning environments, proposed in an earlier work. We have compared the nearest neighbor classifier using the Euclidean and Mahalanobis distance measures with some other classifiers, such as neural networks (MLP and SOM), k-nearest neighbor, Bayesian classifier, Support Vector Machines (SVM), and Optimum Path Forest (OPF). We have also evaluated the Hidden Markov Model (HMM) approach, as a way of using the temporal information. In the experiments with Recogna Video Database, created especially for this study, and Honda/UCSD Video Database, the classifiers obtained the best accuracy, especially the SVM classifier from the SVM Torch library. HMM, which takes into account temporal information, presented better performance than the distance metrics, but worse than the classifiers

Submitted by (ana.araujo@ufrpe.br) on 2016-08-10T13:59:35Z No. of bitstreams: 1 Rita de Cassia de Lima Idalino.pdf: 2600123 bytes, checksum: 41f878b68e3437742821d874a6955502 (MD5) / Made available in DSpace on 2016-08-10T13:59:35Z (GMT). No. of bitstreams: 1 Rita de Cassia de Lima Idalino.pdf: 2600123 bytes, checksum: 41f878b68e3437742821d874a6955502 (MD5) Previous issue date: 2010-12-20 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / Given the large amount of data that is generated in the field of molecular genetics, is of paramount importance that techniques which allow the organization and interpretation of such data be developed and widely disseminated. Initially, we carried out a composition analysis of three gene sequences of the species: ox (Bos taurus), goat (Capra hircus), and sheep (Ovis aries), then we applied alignment techniques for identification of similarities between them. Subsequently, we used the Markov Chain theory with hidden states, i.e. Hidden Markov Models (HMMs, hereafter), in the application of discrimination problem of homogeneous regions in DNA sequences. We used the Viterbi algorithm as an auxiliary tool to obtain homogeneous regions, and then the Baum-Eelch algorithm in order to maximize the probability of a sequence of observations. We analyzed portions of HSP70.1 and NRAMP-1 genes for three different species. / Diante da grande massa de dados que é gerada na área da genética molecular, é de suma importância que técnicas que possibilitem a organização e interpretação desses dados sejam desenvolvidas e amplamente divulgadas. Inicialmente, neste trabalho, foi realizada uma análise da composição de três sequências genéticas, das espécies Bovina (Bos taurus), Caprina (Capra hircus) e Ovina (Ovis aries), em seguida aplicamos técnicas de alinhamentos para identificação de similaridades entre estas. Posteriormente, utilizamos a teoria das cadeias de Markov com estados ocultos, HMM’s (Hidden Markov Models), na aplicação do problema de discriminação de regiões homogêneas em sequências de DNA. Utilizamos o algoritmo de Viterbi como uma ferramenta auxiliar para obtenção de regiões homogêneas e em seguida o algoritmo Baum-Welch para maximizar as probabilidades de uma sequência de observações. Foram analisados trechos dos genes HSP70.1 e NRAMP-1 para três espécies diferentes.

Orientador: Fabio Violaro / Tese (doutorado) - Universidade Estadual de Campinas, Faculdade de Engenharia Eletrica e de Computação / Made available in DSpace on 2018-08-06T10:44:21Z (GMT). No. of bitstreams: 1 Yared_GlaucoFerreiraGazel_D.pdf: 5774867 bytes, checksum: 49a79d9495ce25c8a69ca34858a956ee (MD5) Previous issue date: 2006 / Resumo: Atualmente os sistemas de reconhecimento de fala baseados em HMMs são utilizados em diversas aplicações em tempo real, desde telefones celulares até automóveis. Nesse contexto, um aspecto importante que deve ser considerado é a complexidade dos HMMs, a qual está diretamente relacionada com o custo computacional. Assim, no intuito de permitir a aplicação prática do sistema, é interessante otimizar a complexidade dos HMMs, impondo-se restrições em relação ao desempenho no reconhecimento. Além disso, a otimização da topologia é importante para uma estimação confiável dos parâmetros dos HMMs. Os trabalhos anteriores nesta área utilizam medidas de verossimilhança para a obtenção de sistemas que apresentem um melhor compromisso entre resolução acústica e robustez. Este trabalho apresenta o novo Algoritmo para Eliminação de Gaussianas (GEA), o qual é baseado em uma análise discriminativa e em uma análise interna, para a determinação da complexidade mais apropriada para os HMMs. O novo método é comparado com o Critério de Informação Bayesiano (BIC), com um método baseado em medidas de entropia, com um método discriminativo para o aumento da resolução acústica dos modelos e com os sistemas contendo um número fixo de Gaussianas por estado / Abstract: Nowadays, HMM-based speech recognition systems are used in many real time processing applications, from cell phones to auto mobile automation. In this context, one important aspect to be considered is the HMM complexity, which directly determines the system computational load. So, in order to make the system feasible for practical purposes, it is interesting to optimize the HMM size constrained to a minimum acceptable recognition performance. Furthermore, topology optimization is also important for reliable parameter estimation. Previous works in this area have used likelihood measures in order to obtain models with a better compromise between acoustic resolution and robustness. This work presents the new Gaussian Elimination Algorithm (GEA), which is based on a discriminative analysis and on an internal analysis, for determining the more suitable HMM complexity. The new approach is compared to the classical Bayesian Information Criterion (BIC), to an entropy based method, to a discriminative-based method for increasing the acoustic resolution of the HMMs and also to systems containing a fixed number of Gaussians per state / Doutorado / Telecomunicações e Telemática / Doutor em Engenharia Elétrica

Le travail de recherche de cette thèse de doctorat s'inscrit dans le cadre du suivi médical des patients atteints de démences liées à l'âge à l'aide des caméras videos portées par les patients. L'idée est de fournir aux médecins un nouvel outil pour le diagnostic précoce de démences liées à l'âge telles que la maladie d'Alzheimer. Plus précisément, les Activités Instrumentales du Quotidien (IADL: Instrumental Activities of Daily Living en anglais) doivent être indexées automatiquement dans les vidéos enregistrées par un dispositif d'enregistrement portable.Ces vidéos présentent des caractéristiques spécifiques comme de forts mouvements ou de forts changements de luminosité. De plus, la tâche de reconnaissance visée est d'un très haut niveau sémantique. Dans ce contexte difficile, la première étape d'analyse est la définition d'un équivalent à la notion de « plan » dans les contenus vidéos édités. Nous avons ainsi développé une méthode pour le partitionnement d'une vidéo tournée en continu en termes de « points de vue » à partir du mouvement apparent.Pour la reconnaissance des IADL, nous avons développé une solution selon le formalisme des Modèles de Markov Cachés (MMC). Un MMC hiérarchique à deux niveaux a été introduit, modélisant les activités sémantiques ou des états intermédiaires. Un ensemble complexe de descripteurs (dynamiques, statiques, de bas niveau et de niveau intermédiaire) a été exploité et les espaces de description joints optimaux ont été identifiés expérimentalement.Dans le cadre de descripteurs de niveau intermédiaire pour la reconnaissance d'activités nous nous sommes particulièrement intéressés aux objets sémantiques que la personne manipule dans le champ de la caméra. Nous avons proposé un nouveau concept pour la description d'objets ou d'images faisant usage des descripteurs locaux (SURF) et de la structure topologique sous-jacente de graphes locaux. Une approche imbriquée pour la construction des graphes où la même scène peut être décrite par plusieurs niveaux de graphes avec un nombre de nœuds croissant a été introduite. Nous construisons ces graphes par une triangulation de Delaunay sur des points SURF, préservant ainsi les bonnes propriétés des descripteurs locaux c'est-à-dire leur invariance vis-à-vis de transformations affines dans le plan image telles qu'une rotation, une translation ou un changement d'échelle.Nous utilisons ces graphes descripteurs dans le cadre de l'approche Sacs-de-Mots-Visuels. Le problème de définition d'une distance, ou dissimilarité, entre les graphes pour la classification non supervisée et la reconnaissance est nécessairement soulevé. Nous proposons une mesure de dissimilarité par le Noyau Dépendant du Contexte (Context-Dependent Kernel: CDK) proposé par H. Sahbi et montrons sa relation avec la norme classique L2 lors de la comparaison de graphes triviaux (les points SURF).Pour la reconnaissance d'activités par MMC, les expériences sont conduites sur le premier corpus au monde de vidéos avec caméra portée destiné à l'observation des d'IADL et sur des bases de données publiques comme SIVAL et Caltech-101 pour la reconnaissance d'objets. / The research of this PhD thesis is fulfilled in the context of wearable video monitoring of patients with aged dementia. The idea is to provide a new tool to medical practitioners for the early diagnosis of elderly dementia such as the Alzheimer disease. More precisely, Instrumental Activities of Daily Living (IADL) have to be indexed in videos recorded with a wearable recording device.Such videos present specific characteristics i.e. strong motion or strong lighting changes. Furthermore, the tackled recognition task is of a very strong semantics. In this difficult context, the first step of analysis is to define an equivalent to the notion of “shots” in edited videos. We therefore developed a method for partitioning continuous video streams into viewpoints according to the observed motion in the image plane.For the recognition of IADLs we developed a solution based on the formalism of Hidden Markov Models (HMM). A hierarchical HMM with two levels modeling semantic activities or intermediate states has been introduced. A complex set of features (dynamic, static, low-level, mid-level) was proposed and the most effective description spaces were identified experimentally.In the mid-level features for activities recognition we focused on the semantic objects the person manipulates in the camera view. We proposed a new concept for object/image description using local features (SURF) and the underlying semi-local connected graphs. We introduced a nested approach for graphs construction when the same scene can be described by levels of graphs with increasing number of nodes. We build these graphs with Delaunay triangulation on SURF points thus preserving good properties of local features i.e. the invariance with regard to affine transformation of image plane: rotation, translation and zoom.We use the graph features in the Bag-of-Visual-Words framework. The problem of distance or dissimilarity definition between graphs for clustering or recognition is obviously arisen. We propose a dissimilarity measure based on the Context Dependent Kernel of H. Sahbi and show its relation with the classical entry-wise norm when comparing trivial graphs (SURF points).The experiments are conducted on the first corpus in the world of wearable videos of IADL for HMM based activities recognition, and on publicly available academic datasets such as SIVAL and Caltech-101 for object recognition.

Dans cette thèse, j'étudie les propriétés théoriques des modèles de Markov cachés non paramétriques. Le choix de modèles non paramétriques permet d'éviter les pertes de performance liées à un mauvais choix de paramétrisation, d'où un récent intérêt dans les applications. Dans une première partie, je m'intéresse à l'estimation du nombre d'états cachés. J'y introduis deux estimateurs consistants : le premier fondé sur un critère des moindres carrés pénalisés, le second sur une méthode spectrale. Une fois l'ordre connu, il est possible d'estimer les autres paramètres. Dans une deuxième partie, je considère deux estimateurs adaptatifs des lois d'émission, c'est-à-dire capables de s'adapter à leur régularité. Contrairement aux méthodes existantes, ces estimateurs s'adaptent à la régularité de chaque loi au lieu de s'adapter seulement à la pire régularité. Dans une troisième partie, je me place dans le cadre mal spécifié, c'est-à-dire lorsque les observations sont générées par une loi qui peut ne pas être un modèle de Markov caché. J'établis un contrôle de l'erreur de prédiction de l'estimateur du maximum de vraisemblance sous des conditions générales d'oubli et de mélange de la vraie loi. Enfin, j'introduis une variante non homogène des modèles de Markov cachés : les modèles de Markov cachés avec tendances, et montre la consistance de l'estimateur du maximum de vraisemblance. / During my PhD, I have been interested in theoretical properties of nonparametric hidden Markov models. Nonparametric models avoid the loss of performance coming from an inappropriate choice of parametrization, hence a recent interest in applications. In a first part, I have been interested in estimating the number of hidden states. I introduce two consistent estimators: the first one is based on a penalized least squares criterion, and the second one on a spectral method. Once the order is known, it is possible to estimate the other parameters. In a second part, I consider two adaptive estimators of the emission distributions. Adaptivity means that their rate of convergence adapts to the regularity of the target distribution. Contrary to existing methods, these estimators adapt to the regularity of each distribution instead of only the worst regularity. The third part is focussed on the misspecified setting, that is when the observations may not come from a hidden Markov model. I control of the prediction error of the maximum likelihood estimator when the true distribution satisfies general forgetting and mixing assumptions. Finally, I introduce a nonhomogeneous variant of hidden Markov models : hidden Markov models with trends, and show that the maximum likelihood estimators of such models is consistent.