Centre de documentation
scientifique

Learning to represent and reconstruct 3D deformable objects / Jan Bednarik (2022)  

Public Titre : Learning to represent and reconstruct 3D deformable objects Type de document : Thèse/HDR Auteurs : Jan Bednarik, Auteur ; Pascal Fua, Directeur de thèse ; M. Salzmann, Directeur de thèse Editeur : Lausanne : Ecole Polytechnique Fédérale de Lausanne EPFL Année de publication : 2022 Importance : 138 p. Format : 21 x 30 cm Note générale : bibliographie Thèse présentée pour l'obtention du grade de Docteur ès Sciences, Ecole Polytechnique Fédérale de Lausanne Langues : Anglais (eng) Descripteur : [Vedettes matières IGN] Traitement d'image optique [Termes IGN] appariement de formes [Termes IGN] apprentissage profond [Termes IGN] cohérence temporelle [Termes IGN] déformation de surface [Termes IGN] distorsion d'image [Termes IGN] géométrie de Riemann [Termes IGN] image 3D [Termes IGN] reconstruction d'objet [Termes IGN] semis de points [Termes IGN] vision par ordinateur Index. décimale : THESE Thèses et HDR Résumé : (auteur) Representing and reconstructing 3D deformable shapes are two tightly linked problems that have long been studied within the computer vision field. Deformable shapes are truly ubiquitous in the real world, whether be it specific object classes such as humans, garments and animals or more abstract ones such as generic materials deforming under stress caused by an external force. Truly practical computer vision algorithms must be able to understand the shapes of objects in the observed scenes to unlock the wide spectrum of much sought after applications ranging from virtual try-on to automated surgeries. Automatic shape reconstruction, however, is known to be an ill-posed problem, especially in the common scenario of a single image input. Therefore, the modern approaches rely on deep learning paradigm which has proven to be extremely effective even for the severely under-constrained computer vision problems. We, too, exploit the success of data-driven approaches, however, we also show that generic deep learning models can greatly benefit from being combined with explicit knowledge originating in traditional computational geometry. We analyze the use of various 3D shape representations for deformable object reconstruction and we distinctly focus on one of them, the atlas-based representation, which turns out to be especially suitable for modeling deformable shapes and which we further improve and extend to yield higher quality reconstructions. The atlas-based representation models the surfaces as an ensemble of continuous functions and thus allows for arbitrary resolution and analytical surface analysis. We identify major shortcomings of the base formulation, namely the infamous phenomena of patch collapse, patch overlap and arbitrarily strong mapping distortions, and we propose novel regularizers based on analytically computed properties of the reconstructed surfaces. Our approach counteracts the aforementioned drawbacks while yielding higher reconstruction accuracy in terms of surface normals on the tasks of single view-reconstruction, shape completion and point cloud auto-encoding. We dive into the problematics of atlas-based shape representation even deeper and focus on another pressing design flaw, the global inconsistency among the individual mappings. While the inconsistency is not reflected in the traditional reconstruction accuracy quantitative metrics, it is detrimental to the visual quality of the reconstructed surfaces. Specifically, we design loss functions encouraging intercommunication among the individual mappings which pushes the resulting surface towards a C1 smooth function. Our experiments on the tasks of single-view reconstruction and point cloud auto-encoding reveal that our method significantly improves the visual quality when compared to the baselines. Furthermore, we adapt the atlas-based representation and the related training procedure so that it could model a full sequence of a deforming object in a temporally-consistent way. In other words, the goal is to produce such reconstruction where each surface point always represents the same semantic point on the target ground-truth surface. To achieve such behavior, we note that if each surface point deforms close-to-isometrically, its semantic location likely remains unchanged. Practically, we make use of the Riemannian metric which is computed analytically on the surfaces, and force it to remain point-wise constant throughout the sequence. Our experimental results reveal that our method yields state-of-the-art results on the task of unsupervised dense shape correspondence estimation, while also improving the visual reconstruction quality. Finally, we look into a particular problem of monocular texture-less deformable shape reconstruction, an instance of the Shape-from-Shading problem. We propose a multi-task learning approach which takes an RGB image of an unknown object as the input and jointly produces a normal map, a depth map and a mesh corresponding to the observed part of the surface. We show that forcing the model to produce multiple different 3D representations of the same objects results in higher reconstruction quality. To train the network, we acquire a large real-world annotated dataset of texture-less deforming objects and we release it for public use. Finally, we prove through experiments that our approach outperforms a previous optimization based method on the single-view-reconstruction task. Note de contenu : 1- Introduction 2- Related work 3- Atlas-based representation for deformable shape reconstruction 4- Shape reconstruction by learning differentiable surface representations 5- Better patch stitching for parametric surface reconstruction 6- Temporally-consistent surface reconstruction using metrically-consistent atlases 7- Learning to reconstruct texture-less deformable surfaces from a single view 8- Conclusion Numéro de notice : 15761 Affiliation des auteurs : non IGN Thématique : IMAGERIE Nature : Thèse étrangère Note de thèse : Thèse de Doctorat : Sciences : Lausanne, EPFL : 2022 DOI : 10.5075/epfl-thesis-7974 En ligne : https://doi.org/10.5075/epfl-thesis-7974 Format de la ressource électronique : URL Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=100958

Forest height estimation from a robust TomoSAR method in the case of small tomographic aperture with airborne dataset at L-band / Xing Peng in Remote sensing, vol 13 n° 11 (June-1 2021)  

Public [article]  inRemote sensing > vol 13 n° 11 (June-1 2021) . - n° 2147 Titre : Forest height estimation from a robust TomoSAR method in the case of small tomographic aperture with airborne dataset at L-band Type de document : Article/Communication Auteurs : Xing Peng, Auteur ; Xinwu Li, Auteur ; Yanan Du, Auteur ; et al., Auteur Année de publication : 2021 Article en page(s) : n° 2147 Note générale : bibliographie Langues : Anglais (eng) Descripteur : [Vedettes matières IGN] Traitement d'image radar et applications [Termes IGN] bande L [Termes IGN] données localisées 3D [Termes IGN] forêt boréale [Termes IGN] hauteur des arbres [Termes IGN] image 3D [Termes IGN] image radar moirée [Termes IGN] inventaire forestier (techniques et méthodes) [Termes IGN] itération [Termes IGN] matrice de covariance [Termes IGN] modèle numérique de surface de la canopée [Termes IGN] semis de points [Termes IGN] Suède [Termes IGN] tomographie radar Résumé : (auteur) Forest height is an essential input parameter for forest biomass estimation, ecological modeling, and the carbon cycle. Tomographic synthetic aperture radar (TomoSAR), as a three-dimensional imaging technique, has already been successfully used in forest areas to retrieve the forest height. The nonparametric iterative adaptive approach (IAA) has been recently introduced in TomoSAR, achieving a good compromise between high resolution and computing efficiency. However, the performance of the IAA algorithm is significantly degraded in the case of a small tomographic aperture. To overcome this shortcoming, this paper proposes the robust IAA (RIAA) algorithm for SAR tomography. The proposed approach follows the framework of the IAA algorithm, but also considers the noise term in the covariance matrix estimation. By doing so, the condition number of the covariance matrix can be prevented from being too large, improving the robustness of the forest height estimation with the IAA algorithm. A set of simulated experiments was carried out, and the results validated the superiority of the RIAA estimator in the case of a small tomographic aperture. Moreover, a number of fully polarimetric L-band airborne tomographic SAR images acquired from the ESA BioSAR 2008 campaign over the Krycklan Catchment, Northern Sweden, were collected for test purposes. The results showed that the RIAA algorithm performed better in reconstructing the vertical structure of the forest than the IAA algorithm in areas with a small tomographic aperture. Finally, the forest height was estimated by both the RIAA and IAA TomoSAR methods, and the estimation accuracy of the RIAA algorithm was 2.01 m, which is more accurate than the IAA algorithm with 3.25 m. Numéro de notice : A2021-441 Affiliation des auteurs : non IGN Thématique : FORET/IMAGERIE Nature : Article DOI : 10.3390/rs13112147 Date de publication en ligne : 29/05/2021 En ligne : https://doi.org/10.3390/rs13112147 Format de la ressource électronique : URL article Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=97828 [article]

Benefiting from local rigidity in 3D point cloud processing / Zan Gojcic (2021)  

Public Titre : Benefiting from local rigidity in 3D point cloud processing Type de document : Thèse/HDR Auteurs : Zan Gojcic, Auteur Editeur : Zurich : Eidgenossische Technische Hochschule ETH - Ecole Polytechnique Fédérale de Zurich EPFZ Année de publication : 2021 Importance : 141 p. Format : 21 x 30 cm Note générale : bibliographie A thesis submitted to attain the degree of Doctor of Sciences of ETH Zurich Langues : Français (fre) Descripteur : [Vedettes matières IGN] Lasergrammétrie [Termes IGN] apprentissage profond [Termes IGN] capteur actif [Termes IGN] champ vectoriel [Termes IGN] déformation d'image [Termes IGN] données lidar [Termes IGN] effondrement de terrain [Termes IGN] enregistrement de données [Termes IGN] filtrage du bruit [Termes IGN] flux [Termes IGN] image 3D [Termes IGN] navigation autonome [Termes IGN] orientation du capteur [Termes IGN] segmentation [Termes IGN] semis de points [Termes IGN] téléphone intelligent [Termes IGN] traitement de semis de points [Termes IGN] voxel Index. décimale : THESE Thèses et HDR Résumé : (auteur) Incorporating 3D understanding and spatial reasoning into (intelligent) algorithms is crucial for solving several tasks in fields such as engineering geodesy, risk assessment, and autonomous driving. Humans are capable of reasoning about 3D spatial relations even from a single 2D image. However, making the priors that we rely on explicit and integrating them into computer programs is very challenging. Operating directly on 3D input data, such as 3D point clouds, alleviates the need to lift 2D data into a 3D representation within the task-specific algorithm and hence reduces the complexity of the problem. The 3D point clouds are not only a better-suited input data representation, but they are also becoming increasingly easier to acquire. Indeed, nowadays, LiDAR sensors are even integrated into consumer devices such as mobile phones. However, these sensors often have a limited field of view, and hence multiple acquisitions are required to cover the whole area of interest. Between these acquisitions, the sensor has to be moved and pointed in a different direction. Moreover, the world that surrounds us is also dynamic and might change as well. Reasoning about the motion of both the sensor and the environment, based on point clouds acquired in two-time steps, is therfore an integral part of point cloud processing. This thesis focuses on incorporating rigidity priors into novel deep learning based approaches for dynamic 3D perception from point cloud data. Specifically, the tasks of point cloud registration, deformation analysis, and scene flow estimation are studied. At first, these tasks are incorporated into a common framework where the main difference is in the level of rigidity assumptions that are imposed on the motion of the scene or the acquisition sensor. Then, the tasks specific priors are proposed and incorporated into novel deep learning architectures. While the global rigidity can be assumed in point cloud registration, the motion patterns in deformation analysis and scene flow estimation are more complex. Therefore, the global rigidity prior has to be relaxed to local or instancelevel rigidity, respectively. Rigidity priors not only add structure to the aforementioned tasks, which prevents physically implausible estimates and improves the generalization of the algorithms, but in some cases also reduce the supervision requirements. The proposed approaches were quantitatively and qualitatively evaluated on several datasets, and they yield favorable performance compared to the state-of-the-art. Numéro de notice : 28660 Affiliation des auteurs : non IGN Thématique : IMAGERIE Nature : Thèse étrangère Note de thèse : PhD : Sciences : ETH Zurich : 2021 DOI : sans En ligne : https://www.research-collection.ethz.ch/handle/20.500.11850/523368 Format de la ressource électronique : URL Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=99817

Cartographie dense et compacte par vision RGB-D pour la navigation d’un robot mobile / Bruce Canovas (2021)  

Public Titre : Cartographie dense et compacte par vision RGB-D pour la navigation d’un robot mobile Type de document : Thèse/HDR Auteurs : Bruce Canovas, Auteur ; Michèle Rombaut, Directeur de thèse Editeur : Grenoble [France] : Université Grenoble Alpes Année de publication : 2021 Importance : 148 p. Format : 21 x 30 cm Note générale : bibliographie Thèse pour obtenir le titre de Docteur en Sciences de l'Université Grenoble Alpes, spécialité Signal, Image, Parole, Télécoms Langues : Français (fre) Descripteur : [Vedettes matières IGN] Informatique [Termes IGN] algorithme ICP [Termes IGN] cartographie et localisation simultanées [Termes IGN] image 3D [Termes IGN] image RVB [Termes IGN] instrument embarqué [Termes IGN] modélisation 3D [Termes IGN] navigation autonome [Termes IGN] Ransac (algorithme) [Termes IGN] robot mobile [Termes IGN] vision par ordinateur Index. décimale : THESE Thèses et HDR Résumé : (auteur) Les capacités de perception et de localisation sont des enjeux majeurs de la robotique mobile, nécessaires dans la réalisation de missions impliquant des processus décisionnels autonomes. Elles s'appuient sur les mesures d'un ou plusieurs capteurs embarqués sur la plateforme robotique mobile en question. Dans cette thèse, on s'intéresse à des techniques de perception visuelle basées sur une caméra RGB-D pour permettre la navigation d'un robot compagnon dans un milieu intérieur inconnu. Afin de pouvoir se déplacer de façon autonome, ce robot doit avoir accès à une carte représentative de la structure de son environnement et être capable de s'y repérer.Bien que de nombreuses méthodes de localisation et cartographie simultanées (SLAM - Simultaneous Localization And Mapping) RGB-D capables de résultats impressionnants aient été développées, elles sont bien souvent trop coûteuses en termes de ressources informatiques pour être exécutées sur les cartes embarquées de robots mobiles d'intérieur. Elles font aussi la plupart du temps l'hypothèse que la scène observée par la caméra est statique, ce qui limite leur utilisation dans de nombreuses situations où des personnes sont présentes. De plus, les cartes qu'elles produisent sont souvent inadéquates pour la planification de trajectoires et ne peuvent pas être utilisées directement pour de tâches de navigation.Dans le but de répondre à ces problèmes, nous proposons une nouvelle forme de représentation 3D pour la reconstruction basse résolution, mais compacte rapide et légère d'environnements, au contraire des approches conventionnelles qui se focalisent sur la production de modèles 3D complexes avec un haut niveau détails. Un système de SLAM RGB-D dense complet, robuste aux éléments dynamiques, est conçu autour de cette représentation, puis porté sur une plateforme robotique mobile à conduite différentielle. En outre, une stratégie de navigation efficace est proposée en couplant l'algorithme de SLAM développé à un planificateur de trajectoires. Les différentes solutions proposées sont évaluées et comparées avec les méthodes de l'état de l'art, pour les valider et montrer leurs forces et faiblesses. Note de contenu : 1- Introduction générale 2- Prérequis 3- Modélisation 3D compacte à partir de caméra RGB-D 4- SLAM RGB-D dense, rapide et léger en milieu dynamique 5- Application de SupersurfelFusion à la navigation autonome d’un robot mobile 6- Conclusion et perspectives Numéro de notice : 15283 Affiliation des auteurs : non IGN Thématique : INFORMATIQUE Nature : Thèse française Note de thèse : Thèse de Doctorat : Signal, Image, Parole, Télécoms : Grenoble : 2021 Organisme de stage : Laboratoire Grenoble Images Parole Signal Automatique DOI : sans En ligne : https://tel.hal.science/tel-03647103 Format de la ressource électronique : URL Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=101409

Developing graphics frameworks with Python and OpenGL / Lee Stemkoski (2021)  

Public Titre : Developing graphics frameworks with Python and OpenGL Type de document : Guide/Manuel Auteurs : Lee Stemkoski, Auteur ; Michael Pascale, Auteur Editeur : Boca Raton, New York, ... : CRC Press Année de publication : 2021 Importance : 345 p. Format : 16 x 24 cm ISBN/ISSN/EAN : 978-1-00-318137-8 Langues : Anglais (eng) Descripteur : [Vedettes matières IGN] Langages informatiques [Termes IGN] image 3D [Termes IGN] interface de programmation [Termes IGN] OpenGL [Termes IGN] processeur graphique [Termes IGN] programmation informatique [Termes IGN] Python (langage de programmation) [Termes IGN] scène 3D [Termes IGN] texture d'image [Termes IGN] transformation géométrique Résumé : (éditeur) Developing Graphics Frameworks with Python and OpenGL shows you how to create software for rendering complete three-dimensional scenes. The authors explain the foundational theoretical concepts as well as the practical programming techniques that will enable you to create your own animated and interactive computer-generated worlds. You will learn how to combine the power of OpenGL, the most widely adopted cross-platform API for GPU programming, with the accessibility and versatility of the Python programming language. Topics you will explore include generating geometric shapes, transforming objects with matrices, applying image-based textures to surfaces, and lighting your scene. Advanced sections explain how to implement procedurally generated textures, postprocessing effects, and shadow mapping. In addition to the sophisticated graphics framework you will develop throughout this book, with the foundational knowledge you will gain, you will be able to adapt and extend the framework to achieve even more spectacular graphical results. Note de contenu : 1- Introduction to computer graphics 2- Introduction to Pygame and OpenGL 3- Matrix algebra and transformations 4- A scene graph framework 5- Textures 6- Light and shadow Numéro de notice : 28306 Affiliation des auteurs : non IGN Thématique : INFORMATIQUE Nature : Manuel DOI : 10.1201/9781003181378 En ligne : https://doi.org/10.1201/9781003181378 Format de la ressource électronique : URL Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=98077

Learning-based representations and methods for 3D shape analysis, manipulation and reconstruction / Marie-Julie Rakotosaona (2021)  

Permalink

Intelligent Imaging and Analysis / DaeEun Kim (2020)  

Permalink

Optical Coherence Tomography and Its Non-medical Applications / Michael R. Wang (2020)  

Permalink

Economic losses caused by tree species proportions and site type errors in forest management planning / Arto Haara in Silva fennica, vol 53 n° 2 (2019)  

Permalink

Advanced Remote Sensing Technology for Synthetic Aperture Radar Applications, Tsunami Disasters, and Infrastructure / Maged Marghany (2019)  

Permalink

Close-range photogrammetry and 3D imaging / Thomas Luhmann (2019)  

Permalink

Estimation de profondeur à partir d'images monoculaires par apprentissage profond / Michel Moukari (2019)  

Permalink

Google Earth Engine applications / Lalit Kumar (2019)  

Permalink

Spatial decision support in urban environments using machine learning, 3D geo-visualization and semantic integration of multi-source data / Nikolaos Sideris (2019)  

Permalink

Machine learning and pose estimation for autonomous robot grasping with collaborative robots / Victor Talbot (2018) 

Permalink