Дисертації з теми "SPATIAL TEMPORAL DESCRIPTOR"

L'analyse et l’interprétation de contenus visuels et plus particulièrement la vidéo est un domaine de recherche de plus en plus attractif en raison du nombre important d'applications telles que la vidéo-surveillance, le résumé de films, l'indexation, les jeux vidéo, la robotique et la domotique. Dans cette thèse nous nous intéressons à la détection et à la reconnaissance d'actions humaines dans des séquences vidéo. Pour la partie détection des actions, nous avons introduit deux approches basées sur les points d'intérêts locaux. La première proposition est une méthode simple et efficace qui vise à détecter les mouvements humains ensuite contribuer à extraire des séquences vidéo décrivant des actions importantes. Afin d'atteindre cet objectif, les premières séquences vidéo sont segmentées en volumes de trames et groupes de points d’intérêts. Dans cette méthode, nous nous basons sur le suivi du mouvement des points d'intérêts. Nous avons utilisé, dans un premier lieu, des vidéos simples puis nous avons progressivement augmenté la complexité des vidéos en optant pour des scènes réalistes. Les jeux de données simples présentent généralement un arrière-plan statique avec un Seul acteur qui effectue une seule action unique ou bien la même action mais d'une manière répétitive. Nous avons ensuite testé la robustesse de la détection d'action proposée dans des jeux de données plus complexes réalistes recueillis à partir des réseaux sociaux. Nous avons introduit une approche de détection d'actions efficace pour résoudre le problème de la reconnaissance d'actions humaines dans les vidéos réalistes contenant des mouvements de caméra. Le mouvement humain est donc segmenté d'une manière spatio-temporelle afin de détecter le nombre optimal de trames suffisant pour effectuer une description vidéo. Les séquences sont décrites au moyen de descripteurs spatio-temporels. Nous avons proposé dans cette thèse deux nouveaux descripteurs spatio-temporels basés sur le suivi de la trajectoire des points d'intérêts. Les suivis et la description vidéo sont effectués sur les patchs vidéo qui contiennent un mouvement ou une partie d'un mouvement détecté par la segmentation réalisée lors de l'étape précédente. Nous nous sommes basés sur le descripteur SURF non seulement pour sa précision et mais surtout pour la rapidité. Le premier descripteur proposé est appelé ST-SURF basé sur une nouvelle combinaison du (SURF) et du flot optique. Le ST-SURF permet le suivi de la trajectoire des points d'intérêts tout en gardant les informations spatiales, pertinentes, provenant du SURF. Le deuxième descripteur proposé dans le cadre de cette thèse est un histogramme du mouvement de la trajectoire (HMTO). HMTO est basé sur la position ainsi que l'échelle relative à un SURF. Ainsi, pour chaque SURF détecté, nous définissons une région du voisinage du point d'intérêt en nous basant sur l'échelle. Pour le patch détecté, nous extrayons le flot optique d'une manière dense. Les trajectoires de mouvement sont ensuite générées pour chaque pixel en exploitant les composantes horizontale et verticale de flot optique (u, v). La précision de la description de la vidéo proposée est testée sur un ensemble de données complexes et un plus grand ensemble de données réalistes. Les descripteurs de vidéo proposés sont testés d'une manière simple puis en les fusionnants avec d'autres descripteurs. Les descripteurs vidéo ont été introduits dans un processus de classification basé sur le sac de mots et ont démontré une amélioration des taux de reconnaissance par rapport aux approches précédemment proposés dans l'état-de-l ‘art
Due to increasing demand for video analysis systems in recent years, human action de-tection/recognition is being targeted by the research community in order to make video description more accurate and faster, especially for big datasets. The ultimate purpose of human action recognition is to discern automatically what is happening in any given video. This thesis aims to achieve this purpose by contributing to both action detection and recognition tasks. We thus have developed new description methods for human action recognition.For the action detection component we introduce two novel approaches for human action detection. The first proposition is a simple yet effective method that aims at detecting human movements. First, video sequences are segmented into Frame Packets (FPs) and Group of Interest Points (GIP). In this method we track the movements of Interest Points in simple controlled video datasets and then in videos of gradually increasing complexity. The controlled datasets generally contain videos with a static background and simple ac-tions performed by one actor. The more complex realistic datasets are collected from social networks.The second approach for action detection attempts to address the problem of human ac-tion recognition in realistic videos captured by moving cameras. This approach works by segmenting human motion, thus investigating the optimal sufficient frame number to per-form action recognition. Using this approach, we detect object edges using the canny edge detector. Next, we apply all the steps of the motion segmentation process to each frame. Densely distributed interest points are detected and extracted based on dense SURF points with a temporal step of N frames. Then, optical flows of the detected key points between two frames are computed by the iterative Lucas and Kanade optical flow technique, using pyramids. Since we are dealing with scenes captured by moving cameras, the motion of objects necessarily involves the background and/or the camera motion. Hence, we propose to compensate for the camera motion. To do so, we must first assume that camera motion exists if most points move in the same direction. Then, we cluster optical flow vectors using a KNN clustering algorithm in order to determine if the camera motion exists. If it does, we compensate for it by applying the affine transformation to each frame in which camera motion is detected, using as input parameters the camera flow magnitude and deviation. Finally, after camera motion compensation, moving objects are segmented using temporal differencing and a bounding box is drawn around each detected moving object. The action recognition framework is applied to moving persons in the bounding box. Our goal is to reduce the amount of data involved in motion analysis while preserving the most important structural features. We believe that we have performed action detection in the spatial and temporal domain in order to obtain better action detection and recognition while at the same time considerably reducing the processing time

As advances in the remote sensing of fluid flows forge ahead at an impressive rate, we face an increasingly compelling question of how best to exploit this progress. Light detection and ranging (LIDAR) measurement equipment still presents the problems of having only radial (line-of-sight) wind speed measurements (Cyclops' dilemma). Substantial expanses of unmeasured flow still remain and range weighting errors have a considerable influence on LIDAR measurements. Clearly, more information needs to be extracted from LIDAR data and an estimation problem naturally arises. A key challenge is that most established estimation techniques, such as Kalman filters, cater for systems that are finite-dimensional and described by ordinary differential equations (ODEs). By contrast, many fluid flows are governed by the Navier-Stokes equations, which are nonlinear partial differential-algebraic equations (PDAEs). With this motivation in mind, this thesis proposes a novel statistical signal processing framework for the model-based estimation of a class of spatio-temporal nonlinear partial differential-algebraic equations (PDAEs). The method employs finite-dimensional reduction that converts this formulation to a nonlinear DAE form for which new unscented transform-based filtering and smoothing algorithms are proposed. Gaussian approximations are derived for differential state variables and more importantly, extended to algebraic state variables. A mean-square error lower bound for the nonlinear descriptor filtering problem is obtained based on the posterior Cramér-Rao inequality. The potential of adopting a descriptor systems approach to spatio-temporal estimation is shown for a wind field estimation problem. A basis function decomposition method is used in conjunction with a pressure Poisson equation (PPE) formulation to yield a spatially-continuous, strangeness-free, reduced-order descriptor flow model which is used to estimate unmeasured wind velocities and pressure over the entire spatial region of interest using sparse measurements from wind turbine-mounted LIDAR instruments. The methodology is validated for both synthetic data generated from large eddy simulations of the atmospheric boundary layer and real-world LIDAR measurement data. Results show that a reconstruction of the flow field is achievable, thus presenting a validated estimation framework for potential applications including wind gust prediction systems, the preview control of wind turbines and other spatio-temporal descriptor systems spanning several disciplines.

Lichens are, in most cases, sensitive to anthropogenic factors such as air pollution, global warming, forestry and fragmentation. Two studies are included in this thesis. The first is an evaluation of the importance of old oak for the rare epiphytic lichen Cliostomum corrugatum (Ach.) Fr. This study analysed whether C. corrugatum was limited by dispersal or restricted to tree stands with an unbroken continuity or the substrate old oaks. The results provide evidence that the investigated five populations in Östergötland, Sweden, of C. corrugatum exhibit substantial gene flow, an effective dispersal and a small genetic variation between the sites. Most of the genetic variation was within the populations. Thus, C. corrugatum is more dependent of the substrate old oaks, rather than limited by dispersal. The second study investigated possible range shift of some common macrolichens, due to global warming, from 64 sites in southern Sweden comparing the two years 1986 and 2003. The centroid of three lichen species had moved a significant distance, all in a north east direction: Hypogymnia physodes (L.) Nyl. and Vulpicida pinastri (Scop.) J.-E. Mattsson and M. J. Lai on the tree species Juniperus communis L. (50 and 151 km, respectively) and H. physodes on Pinus sylvestris L. (41 km). Considering also the non-significant cases, there is strong evidence for a prevailing NE direction of centroid movement.

The study presented in this thesis aims to implement and evaluate three different features for the recognition of human activities (HAR). A feature is a piece of information appropriately extracted from raw data that is relevant for solving a particular task. With HAR is meant that discipline able to automatically recognize the activity carried out by an agent through the collection of data. These systems can be realized by exploiting data of different kinds and coming from different sources. Within this study, it was decided to use visual data from properly recorded videos. Each feature is analyzed and evaluated in every detail and finally implemented in a project that uses the most famous and used computer vision and machine learning libraries. The effectiveness of each is evaluated through two different datasets recognized by the scientific community. Each feature was evaluated not only for its performance but also for the computational cost involved in the extraction. The final analysis is therefore the result of a compromise between results obtained and costs to be incurred.

In this thesis, contributions are presented in the areas of shape parsing for view-based object recognition and spatio-temporal feature description for action recognition. A probabilistic model for parsing shapes into several distinguishable parts for accurate shape recognition is presented. This approach is based on robust geometric features that permit high recognition accuracy. As the second contribution in this thesis, a binary spatio-temporal feature descriptor is presented. Recent work shows that binary spatial feature descriptors are effective for increasing the efficiency of object recognition, while retaining comparable performance to state of the art descriptors. An extension of these approaches to action recognition is presented, facilitating huge gains in efficiency due to the computational advantage of computing a bag-of-words representation with the Hamming distance. A scene's motion and appearance is encoded with a short binary string. Exploiting the binary makeup of this descriptor greatly increases the efficiency while retaining competitive recognition performance.

This thesis proposes a new model for describing spatio-temporally deforming objects. Through a novel use of Fourier descriptors, it is shown how arbitrary shape description can be extended to include spatio-temporal shape deformation. It is further demonstrated that these new spatio-temporal Fourier descriptors have the ability to be used as the basis for both the recognition and extraction of deforming objects. Application of this new recognition technique to human gait sequences demonstrates recognition rates of over 86% for individual human subjects, signifying that these descriptors possess unique descriptive properties. Based upon the new spatio-temporal Fourier descriptor model, a new technique for the detection and extraction of deforming shapes from an image sequence is presented through a new variant of the Hough transform (the Continuous Deformable Hough Transform) that utilises spatio-temporal shape correlation within an evidence-gathering context. This new technique demonstrates excellent success rates and tolerance to noise, correctly extracting human subjects in image sequences corrupted with noise levels of up to 80%. The technique is also tested extensively using real-world data, thus demonstrating its worth in a modern-day computer vision system. Both the spatio-temporal Fourier descriptor model, the Continuous Deformable Hough Transform, and aspects of their application are fully discussed throughout the thesis, along with ideas and suggestions for future research and development.

Cette etude se situe dans le domaine de la representation des connaissances et de la formalisation du raisonnement sur l'action. Notre contribution s'interesse particulierement a la localisation d'un mobile a partir d'une description de son itineraire. Nous considerons une scene constituee d'un ensemble de lieux de reference par rapport auxquels un mobile est localise. Deux relations de localisation sont possibles etre a l'interieur ou etre a l'exterieur d'un lieu. L'itineraire du mobile est decrit a travers sa localisation a differents instants, et a travers des actions de deplacements du type entrer ou sortir d'un lieu. Le probleme qui se pose est : que peut-on deduire sur la localisation du mobile a differents instants? deux axes ont ete developpes. Le premier concerne l'etude des problemes de representation de la localisation du mobile par rapport aux lieux, ainsi que la nature spatio-temporelle des deplacements du mobile. Une theorie spatio-temporelle a ainsi ete proposee pour la description de l'evolution d'un mobile par rapport a un ensemble de lieux. Le deuxieme axe developpe concerne l'etude de la representation logique des actions et de leurs effets et particulierement les problemes du decor, de la ramification, du non determinisme et de la concurrence des actions. Une comparaison des differents formalismes dans ce domaine a ete realisee. Nous avons egalement propose un systeme de raisonnement em (modelisation d'evenements), base sur la notion de dependance entre une action et l'etat d'une propriete, et qui permet de resoudre ces differents problemes. Le systeme em constitue la base formelle permettant de decrire les actions de deplacements (entrer et sortir), les lois sur le comportement du mobile, et les hypotheses concernant l'interpretation d'une description. Nous montrons ensuite que les deductions obtenues sont conformes a nos intuitions et au raisonnement de sens commun (suivant la specification de la classe de scenarios etudiee).

L’objectif de cette thèse est de d’écrire l’évolution du produit dans les trois dimensions (spatiale, temporelle et spatio-temporelle). Dans le contexte industriel actuel, les modèles produit sont considérés uniquement du point de vue spatial pendant la phase de conception et du point de vue temporel pendant la phase d’assemblage. Le manque de lien entre le produit et le process mène à des incompréhensions de définition de produit et entraine de mauvaises interprétations en conception. Cependant, le produit ´évolue à travers le temps et subit des changements tout au long des phases de conception et d’assemblage. L’aspect dynamique des activités de conception nécessite de lier ces deux dimensions afin de pouvoir représenter l’évolution du produit et avoir une cohérence des informations. Par conséquent, la dimension spatio-temporelle (i.e. permettant de lier l’espace et le temps) a besoin d’être ajoutée et les relations entre la modélisation du produit et sa séquence d’assemblage ont besoin d’être particulièrement étudiées. Cette thèse en mécanique et conception s’est inspirée de divers domaines comme la gestion des connaissances, les systèmes d’information géographique et la philosophie. Ici le produit est considéré d’un point de vue perdurantiste. Le perdurantisme considère l’objet comme étant compose de tranches temporelles et gardant toujours la même identité quelque soit le changement subi. D’après les précédentes déclarations, cette thèse introduit une nouvelle description du couple produit-process afin d’assurer la compréhension des intentions de conception aux acteurs projet. Dans le but d’atteindre cet objectif, une théorie mereotopologique, permettant de d´écrire le produit comme perçu dans la réalité, et de développée et implémentée dans un modèle ontologique pour être formalisée. La théorie JANUS d´écrit qualitativement l’évolution du produit à travers le temps dans un contexte de conception orientée assemblage, permettant l’intégration de la séquence d’assemblage d`es le début du processus de conception. La théorie permet la description formelle des relations liant les informations et connaissances du couple produit-process. Ces efforts ont pour but d’apporter une base concrète pour la description des changements d’entités spatiales (telles que les composants) et leurs relations `a travers l’espace et le temps. Cette théorie basée sur les régions lie les dimensions spatiale, temporelle et spatio-temporelle et apporte donc une vision perdurantiste en conception de produit. Ensuite, PRONOIA2 – une ontologie formelle basée sur la précédente théorie –développée. De ce fait, les informations liées à l’assemblage sont rendues accessibles et exploitables par des systèmes de gestion d’information et les outils de XAO afin de supporter les activités de l’architecte produit et du concepteur. En effet, les informations et connaissances liées à la conception de produit, ainsi que la séquence d’assemblage associée, ont besoin d’une fondation sémantique et logique afin d’être gérées de manière cohérente et proactive. Suite au développement de la théorie JANUS et de l’ontologie PRONOIA2, l’approche proposée permet d’associer les informations spatiales (gérées par le PDM) et les informations temporelles (gérées par le MPM) à travers des relations mereotopologiques spatio-temporelles. Par conséquent, de nouvelles entités doivent être gérées dans le PLM, en utilisant notamment l’ontologie et un système hub, afin d’assurer un maintien des principes d’ingénierie proactives et améliorer la compréhension de l’architecte produit et du concepteur concernant l’ évolution du produit
The major goal of this research is to describe product evolution in the three dimensions (i.e. spatial, temporal andspatiotemporal). In the current industrial context, product models are only considered from a purely spatial point ofview during the design stage and from a purely temporal point of view during the assembly stage. The lack of linkbetween product and process leads to misunderstanding in engineering definition and causes wrong designinterpretation. However, the product undergoes changes throughout the design and assembly phases. The dynamicaspect of design activities requires linking both dimensions in order to be able to represent product evolution andhave consistent information. As such, spatiotemporal dimension (i.e. linking space and time) needs to be added andrelationships between product modelling and assembly sequences need to be particularly studied.This PhD thesis in mechanical design draws inspiration from several domains such as mathematics, geographicinformation systems and philosophy. Here the product is considered from a perdurantist point of view. Perdurantismregards the object as being composed of temporal slices and always keeping the same identity whatever changesundergone. Based on this statement, this PhD thesis introduces a novel product-process description so as to ensureproduct architect's and designer's understanding of design intents at the early design stages. In order to achieve thisobjective, a mereotopological theory, enabling the product description as it is perceived in the real world, has beendeveloped and implemented in an ontology model to be formalized.The JANUS theory qualitatively describes product evolution over time in the context of AOD, integrating assemblysequence planning in the early product design stages. The theory enables the formal relationships description ofproduct-process design information and knowledge. The proposed efforts aim at providing a concrete basis fordescribing changes of spatial entities (i.e. product parts) and their relationships over time and space. This regionbasedtheory links together spatial, temporal and spatiotemporal dimensions, therefore leading to a perdurantistphilosophy in product design.Then, PRONOIA2 - a formal ontology based on the previous mereotopological theory - is developed. Assemblyinformation is accessible and exploitable by information management systems and computer-aided X tools in orderto support product architects and designer's activities. Indeed product design information and knowledge as well asthe related assembly sequence require a semantic and logical foundation in order to be managed consistently andprocessed proactively.Based on JANUS theory and PRONOIA2 ontology, the MERCURY approach enables associating spatial information(managed by PDM) and temporal information (managed by MPM) through spatiotemporal mereotopologicalrelationships. Therefore, new entities are managed through PLM, using ontology and hub system, so as to ensureproactive engineering and improve product architects' and designers' understanding of product evolution

Avec le développement important des techniques d’animation, les maillages animés sont devenus un sujet de recherche important en informatique graphique, comme la segmentation de maillages animés ou la compression. Ces maillages animés qui sont créés à l’aide de logiciels ou à partir de données de capture de mouvements sont composés d’une séquence ordonnée de maillages de forme statique et dont la topologie reste la même (nombre fixe de sommets et de triangles). Bien qu’un grand nombre de travaux ont été menés sur les maillages statiques durant les deux dernières décennies, le traitement et la compression de maillages animés présentent de nombreuses difficultés techniques. En particulier, les traitements de maillages dynamiques nécessitent une représentation de données efficace basée sur la segmentation. Plusieurs travaux ont été publiés par le passé et qui permettent de segmenter un maillage animé en un ensemble de composants rigides.Dans cette thèse, nous présentons plusieurs techniques qui permettent de calculer une segmentation spatio-temporelle d’un maillage animé ; de tels travaux n’ont pas encore été publiés sur ce sujet. De plus, nous avons étendu cette méthode pour pouvoir comparer ces maillages animés entre eux à l’aide d’une métrique. À notre connaissance, aucune méthode existante ne permet de comparer des maillages animés entre eux
With an abundance of animation techniques available today, animated mesh has become a subject of various data processing techniques in Computer Graphics community, such as mesh segmentation and compression. Created from animation software or from motion capture data, a large portion of the animated meshes are deforming meshes, i.e. ordered sequences of static meshes whose topology is fixed (fixed number of vertices and fixed connectivity). Although a great deal of research on static meshes has been reported in the last two decades, the analysis, retrieval or compressions of deforming meshes remain as new research challenges. Such tasks require efficient representations of animated meshes, such as segmentation. Several spatial segmentation methods based on the movements of each vertex, or each triangle, have been presented in existing works that partition a given deforming mesh into rigid components. In this thesis, we present segmentation techniques that compute the temporal and spatio-temporal segmentation for deforming meshes, which both have not been studied before. We further extend the segmentation results towards the application of motion similarity measurement between deforming meshes. This may be significant as it solves the problem that cannot be handled by current approaches

The chokka squid, Loligo vulgaris reynaudii, is a commercially important species supporting a large, high-value export fishery for South Africa. This study reviewed its population dynamics and reproductive biology. Biological information collected on biannual research cruises from September 1986 to April 1999 on the Agulhas Bank as well as information from commercial vessels operating in the inshore environment collected between April 1988 and July 1989 and again between June 1999 and May 2000 were analysed. Size ranges of juveniles, subadults and adults were calculated and used to determine seasonal distribution and abundance patterns. Based on Gonadosomatic Indices (GSI) and percentages of mature squid in each month, two peak reproductive seasons in summer and winter were identified although mature squid were present all year round. Seasonal peaks in sex ratio were also apparent with males outnumbering females by up to 4:1 in the peak-breeding season. The size at which squid matured, demonstrated both spatial and temporal patterns. Squid caught in spring matured at a smaller size than squid caught in autumn and at successively smaller sizes from west to east across the Agulhas Bank. Size at maturity was highly variable especially in males where “sneaker males” were evident in both seasons. Length-to-weight linear regression revealed significant differences between maturity stages and between sexes. Females demonstrated steeper length-to-weight regression slopes than males in the peak-breeding seasons. Aspects of squid biology pertinent to fishery management were highlighted as well as potential areas where research should be directed in order to develop future stock assessment models.

Cette thèse est une collaboration entre le LAboratoire des Sciences et Matériaux pour l'Électronique et d'Automatique (LASMEA) de Clermont-Ferrand et le Laboratoire Vision et Ingénierie des Contenus (LVIC) du CEA LIST à Saclay. La première moitié de la thèse a été accomplie au sein de l'équipe ComSee (1) du LASMEA et la deuxième au LVIC. L'objectif de ces travaux est de concevoir un système de vidéo-assistance temps réel pour la détection d'évènements dans des scènes possiblement denses.La vidéosurveillance intelligente de scènes denses telles que des foules est particulièrement difficile, principalement à cause de leur complexité et de la grande quantité de données à traiter simultanément. Le but de cette thèse consiste à élaborer une méthode de détection d'évènements rares dans de telles scènes, observées depuis une caméra fixe. La méthode en question s'appuie sur l'analyse automatique de mouvement et ne nécessite aucune information à priori. Les mouvements nominaux sont déterminés grâce à un apprentissage statistique non supervisé. Les plus fréquemment observés sont considérés comme des évènements normaux. Une phase de classification permet ensuite de détecter les mouvements déviant trop du modèle statistique, pour les considérer comme anormaux. Cette approche est particulièrement adaptée aux lieux de déplacements structurés, tels que des scènes de couloirs ou de carrefours routiers. Aucune étape de calibration, de segmentation de l'image, de détection d'objets ou de suivi n'est nécessaire. Contrairement aux analyses de trajectoires d'objets suivis, le coût calculatoire de notre méthode est invariante au nombre de cibles présentes en même temps et fonctionne en temps réel. Notre système s'appuie sur une classification locale du mouvement de la scène, sans calibration préalable. Dans un premier temps, une caractérisation du mouvement est réalisée, soit par des méthodes classiques de flot optique, soit par des descripteurs spatio-temporels. Ainsi, nous proposons un nouveau descripteur spatio-temporel fondé sur la recherche d'une relation linéaire entre les gradients spatiaux et les gradients temporels en des zones où le mouvement est supposé uniforme. Tout comme les algorithmes de flot optique, ce descripteur s'appuie sur la contrainte d'illumination constante.Cependant en prenant en compte un voisinage temporel plus important, il permet une caractérisation du mouvement plus lisse et plus robuste au bruit. De plus, sa faible complexité calculatoire est bien adaptée aux applications temps réel. Nous proposons ensuite d'étudier différentes méthodes de classification : La première, statique, dans un traitement image par image, s'appuie sur une estimation bayésienne de la caractérisation du mouvement au travers d'une approche basée sur les fenêtres de Parzen. Cette nouvelle méthode est une variante parcimonieuse des fenêtres de Parzen. Nous montrons que cette approche est algorithmiquement efficace pour approximer de manière compacte et précise les densités de probabilité. La seconde méthode, basée sur les réseaux bayésiens, permet de modéliser la dynamique du mouvement. Au lieu de considérer ce dernier image par image, des séquences de mouvements sont analysées au travers de chaînes de Markov Cachées. Ajouté à cela, une autre contribution de ce manuscrit est de prendre en compte la modélisation du voisinage d'un bloc afin d'ajouter une cohérence spatiale à la propagation du mouvement. Ceci est réalisé par le biais de couplages de chaînes de Markov cachées.Ces différentes approches statistiques ont été évaluées sur des données synthétiques ainsi qu'en situations réelles, aussi bien pour la surveillance du trafic routier que pour la surveillance de foule.Cette phase d'évaluation permet de donner des premières conclusions encourageantes quant à la faisabilité de la vidéosurveillance intelligente d'espaces possiblement denses.

Fundação de Amparo à Pesquisa do Estado do Rio de Janeiro
Os girinos dos anuros podem ocorrer em inúmeros tipos de sistemas hídricos, desde ambientes relativamente simples e previsíveis, como na água acumulada em epífitas ou uma poça temporária, até hábitats aquáticos permanentes mais complexos, como os riachos. A interação entre os fatores ambientais bióticos e abióticos existentes nesses diferentes ambientes com os fatores históricos é essencial para explicar a estrutura das comunidades dessa fase de vida dos anuros. O entendimento sobre como estes fatores atuam e sua importância nos conduz a uma maior compreensão do que parece influenciar positivamente ou negativamente o estabelecimento dos girinos nos seus diferentes ambientes. Inicialmente, fornecemos uma discussão detalhada da importância desses fatores. Em seguida, avaliamos a estrutura da assembléia dos girinos e a sua estratégia de ocupação espacial e temporal em relação ao uso de diferentes sistemas aquáticos, temporários e permanentes (poças, terrenos alagados, riachos e ambientes artificiais) em uma área de Mata Atlântica na Ilha Grande (Rio de Janeiro). Posteriormente, propomos um experimento para avaliar como os girinos característicos de diferentes tipos de habitats hídricos respondem à condição adversa de ausência de água livre. Depois, é sugerida uma chave artificial de identificação para os girinos da Ilha Grande, com base nas espécies contempladas neste estudo. Por fim, apresentamos a descrição do girino de Proceratophrys tupinamba, provendo algumas informações sobre sua distribuição temporal e uso de microhabitats. Registramos girinos de 12 espécies de anuros, o que correspondeu a 71% dos anfíbios da Ilha Grande com larvas exotróficas em ambientes aquáticos. O espectro de habitats hídricos utilizados variou consistentemente entre as espécies. Girinos de Aplastodiscus eugenioi e Scinax trapicheiroi foram aqueles que utilizaram a maior quantia de tipos de habitats, ambos com cinco registros. A maioria das espécies teve suas maiores abundâncias em um ou dois tipos de corpos dágua onde ocorreu, portanto poucas destas espécies demonstram ter sido generalistas no uso de tipos de habitats aquáticos. A maior riqueza de espécies ocorreu em poças temporárias, em riachos intermitentes e no ambiente antropizado da calha artificial. Quando consideramos em termos de habitats hídricos, a maior riqueza ocorreu nas poças temporárias, nos riachos intermitentes, nos riachos permanentes e na calha artificial. Em nem todos os meses um determinado tipo de recurso hídrico manteve a sua riqueza máxima de girinos. Observamos que um mesmo tipo de sistema hídrico pode comportar espécies típicas de ambientes lênticos e outras adaptadas a ambientes lóticos, dependendo da estrutura em que o corpo dágua apresenta naquele período, como os riachos intermitentes, por exemplo. Entre os fatores abióticos medidos, o PH, o oxigênio dissolvido, a correnteza, a largura e a profundidade dos corpos dágua explicaram de forma mais importante a ocorrência e abundância das diferentes espécies de girinos. Portanto, consideramos que fatores ecológicos desempenham um importante papel na determinação da distribuição de girinos dentro e entre habitats que estes organismos ocupam. O experimento proposto mostrou que os tempos de sobrevivência entre as onze espécies contempladas e também entre os indivíduos de diferentes tamanhos em uma mesma espécie variaram consideravelmente. Isto é sugestivo de que estas espécies apresentam diferentes estratégias para tolerar uma condição de independência de água livre. Os fatores que pareceram mais influenciar negativamente na sobrevivência dos girinos foram: hábito nectônico, pequeno tamanho dos indivíduos, ocupação de ambientes lênticos e temporários e modo reprodutivo não-especializado. Alternativamente, os girinos com melhor desempenho em uma condição de independência de água livre foram de espécies de tamanho comparativamente grande ou médio, ocuparam preferencialmente ambientes lóticos e permanentes, apresentaram modos reprodutivos especializados e os hábitos dos girinos foram principalmente bentônicos. Neste contexto, pode se conjecturar que os girinos das espécies que utilizam ambientes permanentes sejam mais resistentes à condição de independência de água livre do que aquelas de habitats efêmeros. Considerando especial atenção para a biodiversidade dos anfíbios, a Ilha Grande apresenta uma elevada concentração de espécies endêmicas. Esta respeitável diversidade de anfíbios para a área estudada está relacionada com a cobertura vegetal de Mata Atlântica e a grande quantidade de corpos dágua na Ilha, tanto temporários quanto permanentes. A influência destas condições favoráveis para os anfíbios na região está demonstrada também na diversidade de modos reprodutivos, onde 13 dos 39 modos reprodutivos já descritos foram notados para os anfíbios da Ilha Grande. Este conjunto de fatores reafirma esta como uma das mais importantes áreas para a conservação da biodiversidade de anfíbios para o estado do Rio de Janeiro. Comparando a descrição do girino de Proceratophrys tupinamba com P. appendiculata, observamos algumas diferenças na proporção do corpo. Os girinos da espécie descrita foram mais abundantes durante a estação chuvosa (outubro-março), sendo esta distribuição positivamente relacionada com a precipitação média mensal. Os girinos são bentônicos e ocorrem mais frequentemente em porções de menor correnteza do riacho. Eles foram encontrados com maior freqüência expostos na areia, que também representou o microhabitat mais disponível entre aqueles no córrego estudado.
Tadpoles can be found in a number of different aquatic habitats, ranging from relatively simple predictable ecosystems such as bromeliad tank and temporary ponds, to permanent complex ecosystems like streams and rivers. Knowing about the interactions between biotic and abiotic within these habitats is crucial to understand how tadpole communities are structured. Understanding how these factors work and its importance leads to better discerning of what seems positively or negatively influence the establishment of tadpoles in their different environments. Initially, we provide a detailed discussion of the importance of these factors. Next, we evaluate the structure of the assembly of the tadpoles and their strategy of spatial and temporal occupancy in relation to the use of different aquatic systems, temporary and permanent (puddles, flooded land, streams and artificial environments) in an Atlantic Forest area in Ilha Grande (Rio de Janeiro). Subsequently, we propose an experiment to evaluate how tadpoles characteristic of different types of water habitats respond to adverse conditions of absence of free water. Then it is suggested an artificial key to the tadpoles of the Ilha Grande, based on the species covered in this study. Finally, we present the description of the tadpoles of Proceratophrys tupinamba, and provide information on its temporal distribution and microhabitat use. We recorded 12 species of tadpoles, which corresponded to 71% of amphibians of the Ilha Grande with exotrophic larvae in aquatic environments. The spectrum of habitats used varied consistently between species. Tadpoles of Aplastodiscus eugenioi and Scinax trapicheiroi used the greatest amount of habitat types, both with five records. Most species had their highest abundance in one or two types of water bodies where it occurred, so few of these species have been demonstrated to be generalists in the use of aquatic habitats. The highest species richness was found in temporary ponds, in intermittent streams and anthropic environment in the artificial gutter. When considered in terms of water habitats, the richest places were temporary pools, intermittent streams, permanent streams and the artificial gutter. It was not every month that a certain type of water resource maintained its maximum richness of tadpoles. We observe that the same type of water system may include species typical of lentic and lotic environments adapted to other, depending on the structure in which the water body has at that time, such as intermittent streams, for example. Among the abiotic factors measured, the pH, dissolved oxygen, the current, the width and depth of the water bodies more importantly explained the occurrence and abundance of different species of tadpoles. Therefore, we believe that ecological factors play an important role in determining the distribution of tadpoles within and among habitats that these organisms occupy. The proposed experiment showed that the survival times between covered eleven species and between individuals of different sizes in the same species varied considerably. This is suggestive that these species have different strategies to tolerate a condition of independence of free water. The factors that seemed most negative influence on survival of tadpoles were nektonics habit, small individual size, occupation of lentic and temporary habitats and nonspecialized reproductive mode. Alternatively, tadpoles with better performance in a condition of independence of free water were species comparatively large or medium size, that occupied mainly permanent lotic habitats, presented specialized reproductive modes and the tadpoles were mainly benthic. In this context, we can surmise that the tadpoles of species that use permanent environments are more resistant to the condition of independence of free water than those of ephemeral habitats. Noting with particular attention to the biodiversity of amphibians, theIlha Grande has a high concentration of endemic species. This respectable amphibian diversity in the study area is related to the canopy of Atlantic Forest and the large amounts of water bodies on the island, both temporary and permanent. The influence of these favorable conditions for amphibians in the region is also demonstrated in the diversity of reproductive modes, where 13 of the 39 reproductive modes described were noted for the amphibians of the Ilha Grande. These factors reaffirm this as one of the most important areas for biodiversity conservation of amphibians in the state of Rio de Janeiro. Comparing the description of the tadpole of Proceratophrys tupinamba with P. appendiculata, we observed some differences in the proportion of the body. Tadpoles of the species described were more abundant during the rainy season (October to March), this distribution is positively related to average monthly precipitation. The tadpoles are benthic and occur more often in lower stream portions of the creek. They were found more frequently exposed in the sand, which also represented the most microhabitat among those available in the stream studied.

Background: In Africa, the prevalence of HIV/AIDS and hepatitis C virus (HCV) has been the highest in the world. Furthermore, cancer incidence is increasing. Objective: This thesis work presents a part of the impact of these infections on the cancer burden in northern and sub-Saharan Africa. AIDS-related cancers, namely Kaposi sarcoma (KS), non-Hodgkin lymphoma (NHL), and cervical cancer, were described to assess the impact of HIV/AIDS on cancer epidemiology in Algerian, Tunisian, Egyptian, Ugandan, and Zimbabwean populations. Furthermore, for the impact of HCV infection, the epidemiology of NHL as an extra-hepatic HCV-related cancer was also studied. Results: Ugandan HIV/AIDS prevalence has declined since the early 1990s, and in Zimbabwe, after an increase until the end of the 1990s, rates have gone down. In Algeria, the retrospective study performed during this thesis work showed that risk of death and standardized mortality ratio, comparing the mortality of HIV-positive patients with that of the general population, have decreased after the introduction of highly active antiretroviral therapy (HAART) in 1998 in Algeria. However, the late stage of the disease among HIV/AIDS patients at diagnosis warns of the high risk of HIV/AIDS transmission. These observations suggest that Algeria needs to couple HAART use with a more effective prevention programme to fight the increase in the number of HIV/AIDS cases. KS incidence was higher in the sub-Saharan African populations studied than in the northern African populations; however, among women it was similar in Uganda and Zimbabwe. With the emergence of the HIV/AIDS epidemic, KS incidence increased dramatically in both sub-Saharan African populations studied, and it has followed the HIV/AIDS time trend in Zimbabwe. However, in Uganda although HIV/AIDS prevalence has decreased, KS incidence has remained stable among women and elderly men (>50 years old). The decrease in KS incidence was observed only in young Ugandan men (<50 years old). In Uganda and Zimbabwe, we observed an equal or higher KS incidence in men than in women, whereas HIV/AIDS prevalence was higher among women. In northern Africa, KS incidence has remained low and was similar between the populations studied. Regarding NHL, incidence has been highest among both genders in Egypt, while HIV/AIDS has been rare. In Egypt, NHL has seemed to be attributable to HCV infection. In Algeria, another country with low HIV/AIDS prevalence where NHL incidence has increased, a background investigation of NHL patients showed an HCV prevalence of 5.6%. HCV infection may partly explain NHL emergence, especially among a high-risk group of patients for HCV nosocomial contamination. In eastern Algeria, patients found with NHL and HCV infection were elderly women with primarily nosocomial risk factors for HCV contamination. Cervical cancer followed the north–south HIV/AIDS distribution; however, its incidence varied among northern African populations where HIV/AIDS prevalence was similar. Conclusion: KS, NHL, and cervical cancer epidemiology are related to that of HIV/AIDS; however, other infections – human herpesvirus 8 for KS, HCV for NHL, and human papillomavirus for cervical cancer – have influenced the geographical patterns and the time trends of these cancers

Les activités basées sur l'exploitation de données vidéo se sont développées de manière fulgurante ces dernières années. En effet, non seulement avons-nous assisté à une démocratisation de certaines de ces activités, telles que la vidéo-surveillance, mais également à une diversification importante des applications opérationnelles (e.g. suivi de ressources naturelles, reconnaissance aérienne et bientôt satellite). Cependant, le volume de données vidéo généré est aujourd'hui astronomique et l'efficacité des activités correspondantes est limitée par le coût et la durée nécessaire à l'interprétation humaine de ces données vidéo. Par conséquent, l'analyse automatique de flux vidéos est devenue une problématique cruciale pour de nombreuses applications. Les travaux réalisés dans le cadre de cette thèse s'inscrivent dans ce contexte, et se concentrent plus spécifiquement sur l'analyse automatique de vidéos aériennes. En effet, outre le problème du volume de données, ce type de vidéos est particulièrement difficile à exploiter pour un analyste image, du fait des variations de points de vue, de l'étroitesse des champs de vue, de la mauvaise qualité des images, etc. Pour aborder ces difficultés, nous avons choisi de nous orienter vers un système semi-automatique permettant d'assister l'analyste image dans sa tâche, en suggérant des zones d'intérêt potentiel par détection de changements. Plus précisément, l'approche développée dans le cadre de cette thèse cherche à exploiter les données disponibles au maximum de leur potentiel, afin de minimiser l'effort requis pour l'utilisateur et de maximiser les performances de détection. Pour cela, nous effectuons une modélisation tridimensionnelle des apparences observées dans les vidéos de référence. Cette modélisation permet ensuite d'effectuer une détection en ligne des changements significatifs dans une nouvelle vidéo, en identifiant les déviations d'apparence par rapport aux modèles de référence. Des techniques spécifiques ont également été proposées pour effectuer l'estimation des paramètres d'acquisition ainsi que l'atténuation des effets de l'illumination. De plus, nous avons développé plusieurs techniques de consolidation permettant d'exploiter la connaissance a priori relative aux changements à détecter. L'intérêt de notre approche de détection de changements est démontré dans ce manuscrit de thèse, par la présentation des résultats issus de son évaluation minutieuse et systématique. Cette évaluation a été effectuée à l'aide de données réelles et synthétiques permettant d'analyser, d'une part la robustesse de l'approche par rapport à des perturbations réalistes (e.g. bruit, artefacts de compression, apparences et effets complexes, etc), et d'autre part la précision des résultats en conditions contrôlées.

This thesis presents the framework of the combination of motion and shape information produced by different actions for human action recognition. The motion information obtained from the Radon Transform, computed on the Binary Silhouettes obtained from video sequence. Radon Transform of an image gives the projection of lines in all directions therefore it gives information about pixels variation inside the shape. We use the properties of Radon transform i.e. invariance to scaling and translational, to make noise free and robust model and rotational variance to distinguish the different actions. For the shape information we generate the set of static shapes such as MHI/MEI and AEI. Features are extracted from these models by using Pyramid of Histograms (PHOG), Directional Pixels and 2-D DFT. PHOG represents the concatenation of Histogram of Gradients (Hog) over the sub-regions and will give the global spatial information about the shape representation. Shape feature vectors alone will not give discriminating features to recognize the actions such as “run” and “walk”, “jumping at one place” and jumping forward, therefore we integrate with motion descriptor that provides the angular variations while performing actions. We also present another integrated model with motion descriptor where we use Single action images for the representation of pose of action. Single action images are extracted from the sequence of videos using the Fuzzy inference system. Finally with all the extracted features, we train the system using a Support Vector Machine and K-Nearest Neighbour algorithm to recognize the various actions. Weizmann dataset is used for Evaluation.

碩士
國立成功大學
測量及空間資訊學系碩博士班
96
A successful application of geographical information system must be based on correct geographic data that can precisely simulate the real-world phenomenon. Two fundamental factors, time and location, must be considered while collecting geographic data. Since the geographical phenomena may change with time, the design of GIS data model must be able to appropriately model this dynamic nature, in order to accurately reflect the actual changes in reality and enhance the quality in later decision-making process. On the other hand, recent development of Internet and Web Services has brought tremendous improvement to the data sharing mechanism. Easier and instantaneous data acquisition via internet has inspired the motivation of the real-time data application. We argued that it is necessary to correctly describe the spatio-temporal status of reality and distribute data in open format to ensure the successful distribution and application of heterogeneous geographic data in spatial data infrastructure environment. This research focuses on the modeling and sharing of geographic data from a temporal perspective. After analyzing the change patterns of geographical phenomenon and technological restrictions on data collection, we proposed a common spatial-temporal geographic data description framework based on three basic characteristics of geographical phenomenon, namely, time, location and theme properties. The proposed framework can adapt to the modeling of various spatio-temporal change patterns by flexibly including necessary expanded description. These change patterns are concluded by a decision tree that can guide users to choose the most appropriate spatial-temporal data representation according to the real change status of spatial phenomena. To improve data interoperability during sharing process, distributed geographic data is encoded in Geography Markup Language (GML) foramt, which conforms to the standard of the Open Geospatial Consortium (OGC) and International Organization for Standardization (ISO). Finally, we developed a prototype of map interface with augmented capability to justify feature-based spatio-temporal information. While most of the current WebGIS applications still relies on the overlay of static data layers to illustrate only the spatial status of reality, the result of this research demonstrate it is possible to develop an open and feature-based application environment that can indicate the correct spatio-temporal status of the displayed result.

碩士
國立臺灣大學
電機工程學研究所
103
Human activity recognition has become one of the most important areas of research in computer vision. Without extra markers attached to human body, the interaction between machine and human can be natural in a vision-based recognition system. However, it still remains some challenges such as intra-class variability and inter-class similarity. In order to solve the problems, this thesis presents a novel skeleton-based activity recognition methodology with depth sensors. Due to the variation of execution styles and speed for different individuals, a Gaussian blur mask is applied to model the joint position variation while dynamic time warping (DTW) is employed to align the temporal sequences. Besides encoding the structure, the motion is also characterized. Through recording the joint angle trajectory, the entropy of each joint can be evaluated and then be combined with the projected velocity feature. Moreover, the support vector machine (SVM) is applied to acquire the classification results. In our implementation, two challenging public datasets are used to simulate real situations in our daily living. The experimental results show that the proposed approach is discriminative for human activity recognition and performs better than state-of-the-arts. This approach can benefit to several applications such as human-machine interaction (HMI).

碩士
國立清華大學
電機工程學系
88
In this thesis, we will develop a video query by examples (QBE) operation for video database system based on the analysis and description of the spatio-temporal characteristics of the video sequences. Our system uses the characteristic of the video shot to retrieve similar shots in the database based on the video indices, and we processes the video shots to generate three hierarchical information: color, and motion, and object. First. We use Schunck method to estimate the motion field. Here, we analyze each frame to obtain the color level information, and one of every fifteen frame to obtain the motion information. In the color level analysis, we transform the RGB color space to the HSV color space for each frame in the video shot. Then we use the accumulated color histogram to separate the video shot into a series of clips. Then we analyze the clip color histograms for video indexing. In the motion level analysis, we use the spatial-relationship of pixels and the similarity of their motion vectors to cluster pixels into region, and generate a so-called motion image, which can be used to obtain the local motion histogram and global motion histogram. In the object level analysis, we search for the video objects, which are moving objects in the video sequence. We first use the difference image and the data of the motion field to find out the regions of object. Then we take a region tracing between the sub-sampled frame to find the object regions. Using these traced regions, we can calculate the centroid, and the size, and the average motion vector of the moving objects.