No description available.

Cette thèse présente une approche de segmentation d’images sans supervision pour obtenir une série de silhouettes et l’enveloppe visuelle («Visual Hull») d’un objet observé à partir de plusieurs points de vue. L’approche proposée peut traiter essentiellement n’importe quelles propriétés d’apparence comme la texture, des couleurs similaires d’arrière-plan, la specularité, la transparence et d’autre phénomènes tels que des ombres et des débordements de couleur. Par rapport aux méthodes plus classiques d’extraction de silhouettes sur plusieurs vues, où certaines hypothèses sur l’objet ou la scène sont formulès, nous ne modélisons ni l’arrière-plan ni les propriétés d’apparence de l’objet. La seule hypothèse est la constance de l’arrière-plan considéré comme inconnu pour un point de vue lorsque l’objet bouge. L’idée principale de l’approche est d’estimer l’évolution temporelle de chaque pixel pour obtenir une mesure de stabilité. Celle-ci est utilisée pour construire la fonction de coût d’arrière-plan. Pour faire face aux ombres et aux ombres projetées, un objet est capturé sous différentes conditions d’éclairage. En outre, les informations d’espace, de temps et d’éclairage sont fusionnées et utilisées dans un formalisme de champ aléatoire de Markov et la fonction d’énergie construite est minimisée par «Graph Cuts». Les expérimentations en laboratoire, sous différentes conditions d’éclairage, montrent que l’approche proposée permet la reconstruction robuste de l’enveloppe visuelle pour une grande variété d’objets difficiles tels que les objets en verre (effets de transparence) ou en métal brillant (effets de réflexions spéculaires). / This thesis presents an unsupervised image segmentation approach for obtaining a set of silhouettes along with the Visual Hull of an object observed from multiple viewpoints. The proposed approach can deal with mostly any type of appearance characteristics such as texture, similar background color, shininess, transparency besides other phenomena such as shadows and color bleeding. Compared to more classical methods for silhouette extraction from multiple views, for which certain assumptions are made on the object or scene, neither the background nor the object appearance properties are modeled. The only assumption is the constancy of the unknown background for a given camera viewpoint while the object is under motion. The principal idea of the approach is the estimation of the temporal evolution of each pixel over time which provides a stability measurement and leads to its associated background cost function. In order to cope with shadows and self-shadows, an object is captured under different lighting conditions. Furthermore, the information from the space, time and lighting domains is exploited and merged based on a Markov Random Field framework and the constructed energy function is minimized via graph-cut. Real-data experiments under different lighting condition show that the proposed approach allows for robust Visual Hull reconstruction of a variety of challenging objects such as objects made of shiny metal or glass.

Les silhouettes ambiguës, comme celle du lapin/canard (Jastrow, 1899), ont été étudiées selon plusieurs approches. Toutefois, les figures prises en exemples dans la large majorité des études sont généralement les mêmes. Cette redondance des images ambiguës utilisées pousse à croire qu'elles sont peut-être assez rares. Certaines observations anecdotiques suggèrent cependant qu’elles seraient au contraire relativement fréquentes. C'est ce que cherche à déterminer cette expérience. Nous avons utilisé des modèles tridimensionnels d'animaux projetés de façon aléatoire afin d'en extraire les silhouettes dont la complexité périmétrique a ensuite été modifiée par lissage. Treize sujets ont dû indiquer ce qu'ils percevaient dans l'image. Nous démontrons qu’une silhouette est classée en moyenne dans 1.9079 catégories de base. Nous avons également démontré qu’une diminution de la complexité périmétrique rend d’abord une silhouette plus ambiguë pour éventuellement atteindre un sommet (équivalent à environ six fois la complexité périmétrique d’un disque) à la suite duquel l’ambiguïté chute. / Ambiguous silhouettes such as the duck/rabbit (Jastrow, 1899) have been studied by several approaches. However, the figures taken as examples in the vast majority of studies are generally the same. This redundancy of the ambiguous images used in litterature implies they may be quite rare. On the other hand, anecdotal evidence suggests that they might be relatively frequent. This is what this experiment is trying to establish. We used three-dimensional models of animals from a random point of view to extract silhouettes whose perimetric complexity was subsequently modified by smoothing. Thirteen subjects were asked to indicate what they saw in the image. We show that silhouettes are classified on average with 1.9079 based categories. We also established that a decrease in the perimetric complexity initially makes a more ambiguous figure but that this effect eventually reaches a peak (at a perimetric complexity of approximately 6 times that of a disk) after which ambiguity drops.

Knowing the shape and pose of objects of interest is critical information when planning robotic grasping and manipulation maneuvers. The ability to recover this information from objects for which the system has no prior knowledge is a valuable behavior for an autonomous or semiautonomous robot. This work develops and presents an algorithm for the shape and pose recovery of unknown objects using no a priori information. Using a monocular camera in an eye-in-hand configuration, three images of the object of interest are captured from three disparate viewing directions. Machine vision techniques are employed to process these images into silhouettes. The silhouettes are used to generate an approximation of the surface of the object in the form of a three dimensional point cloud. The accuracy of this approximation is improved by fitting an eleven parameter geometric shape to the points such that the fitted shape ignores disturbances from noise and perspective projection effects. The parametrized shape represents the model of the unknown object and can be utilized for planning robot grasping maneuvers or other object classification tasks. This work is implemented and tested in simulation and hardware. A simulator is developed to test the algorithm for various three dimensional shapes and any possible imaging positions. Several shapes and viewing configurations are tested and the accuracy of the recoveries are reported and analyzed. After thorough testing of the algorithm in simulation, it is implemented on a six axis industrial manipulator and tested on a range of real world objects: both geometric and amorphous. It is shown that the accuracy of the hardware implementation performs exceedingly well and approaches the accuracy of the simulator, despite the additional sources of error and uncertainty present.

Les silhouettes ambiguës, comme celle du lapin/canard (Jastrow, 1899), ont été étudiées selon plusieurs approches. Toutefois, les figures prises en exemples dans la large majorité des études sont généralement les mêmes. Cette redondance des images ambiguës utilisées pousse à croire qu'elles sont peut-être assez rares. Certaines observations anecdotiques suggèrent cependant qu’elles seraient au contraire relativement fréquentes. C'est ce que cherche à déterminer cette expérience. Nous avons utilisé des modèles tridimensionnels d'animaux projetés de façon aléatoire afin d'en extraire les silhouettes dont la complexité périmétrique a ensuite été modifiée par lissage. Treize sujets ont dû indiquer ce qu'ils percevaient dans l'image. Nous démontrons qu’une silhouette est classée en moyenne dans 1.9079 catégories de base. Nous avons également démontré qu’une diminution de la complexité périmétrique rend d’abord une silhouette plus ambiguë pour éventuellement atteindre un sommet (équivalent à environ six fois la complexité périmétrique d’un disque) à la suite duquel l’ambiguïté chute. / Ambiguous silhouettes such as the duck/rabbit (Jastrow, 1899) have been studied by several approaches. However, the figures taken as examples in the vast majority of studies are generally the same. This redundancy of the ambiguous images used in litterature implies they may be quite rare. On the other hand, anecdotal evidence suggests that they might be relatively frequent. This is what this experiment is trying to establish. We used three-dimensional models of animals from a random point of view to extract silhouettes whose perimetric complexity was subsequently modified by smoothing. Thirteen subjects were asked to indicate what they saw in the image. We show that silhouettes are classified on average with 1.9079 based categories. We also established that a decrease in the perimetric complexity initially makes a more ambiguous figure but that this effect eventually reaches a peak (at a perimetric complexity of approximately 6 times that of a disk) after which ambiguity drops.

O presente estudo teve por objetivo avaliar os efeitos ao longo de 30 sessões de mat Pilates em mulheres sedentárias. Foi desenvolvida uma Escala de Imagem Corporal a qual apresentou validade de conteúdo segundo sete experts, uma boa repetibilidade (Kappa = 0,743; p < 0,001) e reprodutibilidade (Kappa = 0,637; p < 0,001), além de uma boa correlação entre a silhueta apontada como a atual e o IMC do indivíduo (r = 0,696). 21 mulheres sedentárias sem contato prévio com o método e idade entre 21 e 35 anos foram submetidas a 30 sessões de Pilates em turmas de quatro alunas, duas vezes por semana. As avaliações foram realizadas em cinco momentos: antes do período controle de um mês, antes da intervenção, após dez sessões, após 20 sessões e após 30 sessões. Foi avaliada a força de preensão palmar como representativa da força geral e de forma complementar a força dos extensores do tronco; para a capacidade cardiorrespiratória foram considerados o primeiro e o segundo limar ventilatório e o VO2; e para a composição corporal o perfil de cinco componentes e o somatório de seis dobras. Foi encontrada diferença significativa e um tamanho de efeito grande na capacidade cardiorrespiratória entre as avaliações pré-intervenção e após trinta sessões: no primeiro limiar ventilatório (r = 0,77; p < 0,05), no segundo limiar ventilatório (r = 0,85; p < 0,05) e também no VO2 (r = 0,81; p < 0,05). A força dos extensores da coluna apresentou diferença estatística e tamanho de efeito grande entre o período pré e pós-intervenção (T = 9,50; r = - 0,537; z = -3,311; p = 0,001). Ao longo de trinta sessões de método Pilates houve diferença apenas nas capacidades cardiorrespiratórias e na força de extensores do tronco. A imagem corporal, a preensão palmar e a composição corporal não foram alteradas. / This study aimed to evaluate the effects along thirty mat Pilates sessions on women. A Body Image Scale was developed, which presented content validity according to seven experts, good repeatability (Kappa = 0.743; p < 0.001) and reproducibility (Kappa = 0.637; p < 0.001), and good correlation between the individual’s current self-identified silhouette and the respective BMI (r = 0.696). 21 women with no prior contact with the method and aged between 21 and 35 years were subjected to 30 Pilates sessions, twice a week, in groups of four participants. The evaluations were conducted at five moments: one month before the initiation of the intervention, immediately before the intervention, after ten sessions, after 20 sessions and after 30 sessions. The evaluation included: hand grip strength, representing overall strength and particularly trunk extensors strength; the first and second ventilatory threshold and VO2, for cardiorespiratory fitness; and a five-component profile and the sum of six skinfolds, for body composition. A significant difference and a large size effect were found on cardiorespiratory fitness between pre intervention and after thirty sessions ratings: first ventilatory threshold (r = 0.77; p < 0.05) in the second ventilatory threshold (r = 0.85; p < 0.05) and also in VO2 (r = 0.81; p < 0.05). The strength of the trunk extensors showed statistical significance and large effect size between pre and post intervention (t = 9.50; r = - 0.537; z = -3.311; p = 0.001). However, no difference in grip strength, body image or composition was found.

O presente estudo teve por objetivo avaliar os efeitos ao longo de 30 sessões de mat Pilates em mulheres sedentárias. Foi desenvolvida uma Escala de Imagem Corporal a qual apresentou validade de conteúdo segundo sete experts, uma boa repetibilidade (Kappa = 0,743; p < 0,001) e reprodutibilidade (Kappa = 0,637; p < 0,001), além de uma boa correlação entre a silhueta apontada como a atual e o IMC do indivíduo (r = 0,696). 21 mulheres sedentárias sem contato prévio com o método e idade entre 21 e 35 anos foram submetidas a 30 sessões de Pilates em turmas de quatro alunas, duas vezes por semana. As avaliações foram realizadas em cinco momentos: antes do período controle de um mês, antes da intervenção, após dez sessões, após 20 sessões e após 30 sessões. Foi avaliada a força de preensão palmar como representativa da força geral e de forma complementar a força dos extensores do tronco; para a capacidade cardiorrespiratória foram considerados o primeiro e o segundo limar ventilatório e o VO2; e para a composição corporal o perfil de cinco componentes e o somatório de seis dobras. Foi encontrada diferença significativa e um tamanho de efeito grande na capacidade cardiorrespiratória entre as avaliações pré-intervenção e após trinta sessões: no primeiro limiar ventilatório (r = 0,77; p < 0,05), no segundo limiar ventilatório (r = 0,85; p < 0,05) e também no VO2 (r = 0,81; p < 0,05). A força dos extensores da coluna apresentou diferença estatística e tamanho de efeito grande entre o período pré e pós-intervenção (T = 9,50; r = - 0,537; z = -3,311; p = 0,001). Ao longo de trinta sessões de método Pilates houve diferença apenas nas capacidades cardiorrespiratórias e na força de extensores do tronco. A imagem corporal, a preensão palmar e a composição corporal não foram alteradas. / This study aimed to evaluate the effects along thirty mat Pilates sessions on women. A Body Image Scale was developed, which presented content validity according to seven experts, good repeatability (Kappa = 0.743; p < 0.001) and reproducibility (Kappa = 0.637; p < 0.001), and good correlation between the individual’s current self-identified silhouette and the respective BMI (r = 0.696). 21 women with no prior contact with the method and aged between 21 and 35 years were subjected to 30 Pilates sessions, twice a week, in groups of four participants. The evaluations were conducted at five moments: one month before the initiation of the intervention, immediately before the intervention, after ten sessions, after 20 sessions and after 30 sessions. The evaluation included: hand grip strength, representing overall strength and particularly trunk extensors strength; the first and second ventilatory threshold and VO2, for cardiorespiratory fitness; and a five-component profile and the sum of six skinfolds, for body composition. A significant difference and a large size effect were found on cardiorespiratory fitness between pre intervention and after thirty sessions ratings: first ventilatory threshold (r = 0.77; p < 0.05) in the second ventilatory threshold (r = 0.85; p < 0.05) and also in VO2 (r = 0.81; p < 0.05). The strength of the trunk extensors showed statistical significance and large effect size between pre and post intervention (t = 9.50; r = - 0.537; z = -3.311; p = 0.001). However, no difference in grip strength, body image or composition was found.

O presente estudo teve por objetivo avaliar os efeitos ao longo de 30 sessões de mat Pilates em mulheres sedentárias. Foi desenvolvida uma Escala de Imagem Corporal a qual apresentou validade de conteúdo segundo sete experts, uma boa repetibilidade (Kappa = 0,743; p < 0,001) e reprodutibilidade (Kappa = 0,637; p < 0,001), além de uma boa correlação entre a silhueta apontada como a atual e o IMC do indivíduo (r = 0,696). 21 mulheres sedentárias sem contato prévio com o método e idade entre 21 e 35 anos foram submetidas a 30 sessões de Pilates em turmas de quatro alunas, duas vezes por semana. As avaliações foram realizadas em cinco momentos: antes do período controle de um mês, antes da intervenção, após dez sessões, após 20 sessões e após 30 sessões. Foi avaliada a força de preensão palmar como representativa da força geral e de forma complementar a força dos extensores do tronco; para a capacidade cardiorrespiratória foram considerados o primeiro e o segundo limar ventilatório e o VO2; e para a composição corporal o perfil de cinco componentes e o somatório de seis dobras. Foi encontrada diferença significativa e um tamanho de efeito grande na capacidade cardiorrespiratória entre as avaliações pré-intervenção e após trinta sessões: no primeiro limiar ventilatório (r = 0,77; p < 0,05), no segundo limiar ventilatório (r = 0,85; p < 0,05) e também no VO2 (r = 0,81; p < 0,05). A força dos extensores da coluna apresentou diferença estatística e tamanho de efeito grande entre o período pré e pós-intervenção (T = 9,50; r = - 0,537; z = -3,311; p = 0,001). Ao longo de trinta sessões de método Pilates houve diferença apenas nas capacidades cardiorrespiratórias e na força de extensores do tronco. A imagem corporal, a preensão palmar e a composição corporal não foram alteradas. / This study aimed to evaluate the effects along thirty mat Pilates sessions on women. A Body Image Scale was developed, which presented content validity according to seven experts, good repeatability (Kappa = 0.743; p < 0.001) and reproducibility (Kappa = 0.637; p < 0.001), and good correlation between the individual’s current self-identified silhouette and the respective BMI (r = 0.696). 21 women with no prior contact with the method and aged between 21 and 35 years were subjected to 30 Pilates sessions, twice a week, in groups of four participants. The evaluations were conducted at five moments: one month before the initiation of the intervention, immediately before the intervention, after ten sessions, after 20 sessions and after 30 sessions. The evaluation included: hand grip strength, representing overall strength and particularly trunk extensors strength; the first and second ventilatory threshold and VO2, for cardiorespiratory fitness; and a five-component profile and the sum of six skinfolds, for body composition. A significant difference and a large size effect were found on cardiorespiratory fitness between pre intervention and after thirty sessions ratings: first ventilatory threshold (r = 0.77; p < 0.05) in the second ventilatory threshold (r = 0.85; p < 0.05) and also in VO2 (r = 0.81; p < 0.05). The strength of the trunk extensors showed statistical significance and large effect size between pre and post intervention (t = 9.50; r = - 0.537; z = -3.311; p = 0.001). However, no difference in grip strength, body image or composition was found.

Nous nous intéressons dans cette thèse à la modélisation de formes tridimensionnelles à partir de vidéos numériques. Plus particulièrement nous considérons les approches utilisant des silhouettes, acquises à l'aide de caméras calibrées. Notre objectif est de fournir une modélisation rapide pour permettre au modèle construit d'interagir en temps réel avec des environnements virtuels. Parmi les différentes primitives qui peuvent être considérées pour acquérir des modèles tridimensionnels, les silhouettes offrent l'avantage de simplifier la gestion de la visibilité et de permettre des traitements rapides. Les principales limitations des méthodes qui en découlent concernent le rapport défavorable entre la précision du modèle construit et le temps de calcul nécessaire, ainsi que l'extraction des silhouettes dans les images qui reste une étape sensible du processus de modélisation. En réponse à ces limitations, nous proposons deux approches rapides de modélisation à partir de silhouettes. La première suppose que les données, les silhouettes et le calibrage, sont précises, et produit un modèle exact vis-à-vis de ces données. La deuxième permet d'introduire des incertitudes dans les primitives considérées et produit alors un modèle probabiliste du volume observé. Ces deux approches sont illustrées par des résultats sur données synthétiques et réelles, acquises sur la plate-forme GrImage de l'Inria Rhône-Alpes. En particulier une implémentation distribuée et temps réel de la première méthode est proposée. Celle-ci est validée dans le contexte d'une application de réalité virtuelle.

La classification sur ombre de mines marines peut être effectuée à partir d'images sonar haute résolution. Cependant, la classification mono-vue admet des limites du fait de leur forme géométrique parfois complexe. Une manière de s'affranchir des ambiguïtés consiste à effectuer une classification multi-vues. Dans un premier temps, il convient d'assimiler les spécificités de l'imagerie sonar. D'un côté, les propriétés statistiques des pixels et le principe d'acquisition des images sont des connaissances avantageusement prises en compte lors de l'étape de segmentation. D'un autre côté, les distorsions géométriques qui affectent l'ombre portée du fait des performances du sonar et de la prise de vue doivent être considérées. En matière de reconnaissance de formes, les données sonar segmentées peuvent être diversement résumées suivant qu'on s'attache à définir des grandeurs indépendantes des positions relatives objet-sonar et/ou de la résolution du sonar ou bien, autorisant une reconstruction du contour et la conservation de l'orientation de la forme associée. Ces considérations du problème mono-vue ont préparé l'élaboration de processus plus complexes de classification multi-vues. D'une part, on manipule les attributs issus d'un traitement image par image. Deux cas se présentent : en exploitant l'évolution de la forme de l'ombre en fonction du trajet du sonar ou, au contraire, en s'affranchissant des transformations du plan. Dans le premier cas, on caractérise de manière globale l'ensemble des valeurs successives prises par des attributs sensibles aux diverses formes de l'ombre. Dans le second cas, sans connaissance précise des conditions d'acquisition, les attributs extraits sont fusionnés et présentés à l'opérateur sous forme de mesures pour l'aide à la décision. Par la logique floue d'autre part, les outils de reconnaissance de formes calculés sur des données binaires ont été étendus au cas de données en niveaux de gris d'une nouvelle image pour sa caractérisation.