L'objectif principal de ce travail est le développement de recherches en modélisation 3D du bâti. En particulier, la recherche en reconstruction 3D de bâtiment est un sujet très développé depuis les années 90. Malgré tout, il paraît nécessaire de poursuivre les recherches dans cet axe étant données que les approches actuelles consacrées à la reconstruction 3D de bâtiment (bien qu'efficaces) rencontrent encore des difficultés en terme de généralisation, de cohérence et de précision. Par ailleurs, les récents développements des systèmes d'acquisitions de rues tel que les systèmes de cartographie mobile ouvrent de nouvelles perspectives d'amélioration de la modélisation des bâtiments dans le sens ou les données terrestres (très précises et résolus) peuvent être exploitées avec davantage de cohérence (en comparaison à l'aérien) pour enrichir les modèles de bâtiments au niveau des façades (la géométrie, la texture).Ainsi, des approches de modélisation aériennes et terrestres sont individuellement proposées. Au niveau aérien, nous décrivons une approche directe et dépourvu d'extraction et d'assemblage de primitives géométriques en vue de la reconstruction 3D de modèles polyédriques simples de bâtiments à partir d'un jeu d'images aériennes calibrées. Au niveau terrestre, plusieurs approches qui décrivent essentiellement un pipeline pour la modélisation 3D des façades urbaines sont proposées; à savoir, la segmentation et classification de nuage de rues urbaines, la modélisation géométrique des façades urbaines et le texturage des façades urbaines comportant des occultations causées par d'autres objets du mobilier urbains / The aim of this work is to develop research on 3D building modeling. In particular, the research in aerial-based 3D building reconstruction is a topic very developed since 1990. However, it is necessary to pursue the research since the actual approaches for 3D massive building reconstruction (although efficient) still encounter problems in generalization, coherency, accuracy. Besides, the recent developments of street acquisition systems such as Mobile Mapping Systems open new perspectives for improvements in building modeling in the sense that the terrestrial data (very dense and accurate) can be exploited with more performance (in comparison to the aerial investigation) to enrich the building models at facade level (e.g., geometry, texturing).Hence, aerial and terrestrial based building modeling approaches are individually proposed. At aerial level, we describe a direct and featureless approach for simple polyhedral building reconstruction from a set of calibrated aerial images. At terrestrial level, several approaches that essentially describe a 3D urban facade modeling pipeline are proposed, namely, the street point cloud segmentation and classification, the geometric modeling of urban facade and the occlusion-free facade texturing

Les robots et véhicules autonomes représentent le futur des modes de déplacements et de production. Les enjeux de l’avenir reposent sur la robustesse de leurs perceptions et flexibilité face aux environnements changeant et situations inattendues. Les capteurs stéréoscopiques sont des capteurs passifs qui permettent d'obtenir à la fois image et information 3D de la scène à la manière de la vision humaine. Dans ces travaux nous avons développé un système de localisation, par odométrie visuelle permettant de déterminer la position dans l'espace du capteur de façon efficace et performante en tirant partie de la carte de profondeur dense mais également associé à un système de SLAM, rendant la localisation robuste aux perturbations et aux décalages potentiels. Nous avons également développé plusieurs solutions de cartographie et interprétation d’obstacles, à la fois pour le véhicule aérien et terrestre. Ces travaux sont en partie intégrés dans des produits commerciaux. / Autonomous vehicles and robots represent the future of transportation and production industries. The challenge ahead will come from the robustness of perception and flexibility from unexpected situations and changing environments. Stereoscopic cameras are passive sensors that provide color images and depth information of the scene by correlating 2 images like the human vision. In this work, we developed a localization system, by visual odometry that can determine efficiently the position in space of the sensor by exploiting the dense depth map. It is also combined with a SLAM system that enables robust localization against disturbances and potentials drifts. Additionally, we developed a few mapping and obstacles detections solutions, both for aerial and terrestrial vehicles. These algorithms are now partly integrated into commercial products.

Orientador: Antonio Maria Garcia Tommaselli / Resumo: A combinação de duas lentes com FOV hiper-hemisférico em posição opostas pode gerar um sistema omnidirecional (FOV 360°) leve, compacto e de baixo custo, como Ricoh Theta S e GoPro Fusion. Entretanto, apenas algumas técnicas e modelos matemáticos para a calibração um sistema com duas lentes hiper-hemisféricas são apresentadas na literatura. Nesta pesquisa, é avaliado e definido um modelo geométrico para calibração de sistemas omnidirecionais compostos por duas lentes hiper-hemisféricas e apresenta-se algumas aplicações com esse tipo de sistema. A calibração das câmaras foi realizada no programa CMC (calibração de múltiplas câmeras) utilizando imagens obtidas a partir de vídeos feitos com a câmara Ricoh Theta S no campo de calibração 360°. A câmara Ricoh Theta S é composto por duas lentes hiper-hemisféricas fisheye que cobrem 190° cada uma. Com o objetivo de avaliar as melhorias na utilização de pontos em comum entre as imagens, dois conjuntos de dados de pontos foram considerados: (1) apenas pontos no campo hemisférico, e (2) pontos em todo o campo de imagem (isto é, adicionar pontos no campo de imagem hiper-hemisférica). Primeiramente, os modelos ângulo equisólido, equidistante, estereográfico e ortogonal combinados com o modelo de distorção Conrady-Brown foram testados para a calibração de um sensor da câmara Ricoh Theta S. Os modelos de ângulo-equisólido e estereográfico apresentaram resultados melhores do que os outros modelos. Portanto, esses dois modelos de projeção for... (Resumo completo, clicar acesso eletrônico abaixo) / Abstract: The arrangement of two hyper-hemispherical fisheye lenses in opposite position can design a light weight, small and low-cost omnidirectional system (360° FOV), e.g. Ricoh Theta S and GoPro Fusion. However, only a few techniques are presented in the literature to calibrate a dual-fisheye system. In this research, a geometric model for dual-fisheye system calibration was evaluated, and some applications with this type of system are presented. The calibrating bundle adjustment was performed in CMC (calibration of multiple cameras) software by using the Ricoh Theta video frames of the 360° calibration field. The Ricoh Theta S system is composed of two hyper-hemispherical fisheye lenses with 190° FOV each one. In order to evaluate the improvement in applying points in the hyper-hemispherical image field, two data set of points were considered: (1) observations that are only in the hemispherical field, and (2) points in all image field, i.e. adding points in the hyper-hemispherical image field. First, one sensor of the Ricoh Theta S system was calibrated in a bundle adjustment based on the equidistant, equisolid-angle, stereographic and orthogonal models combined with Conrady-Brown distortion model. Results showed that the equisolid-angle and stereographic models can provide better solutions than those of the others projection models. Therefore, these two projection models were implemented in a simultaneous camera calibration, in which the both Ricoh Theta sensors were considered i... (Complete abstract click electronic access below) / Mestre

Un défi central en robotique est la construction de représentations du monde exploitables. Pour accomplir des tâches complexes, les robots doivent construire une représentation 3D de leur environnement qui représente les informations géométriques, visuelles et sémantiques de la scène, et qui est efficace à utiliser. Les approches existantes encodent les informations sémantiques en utilisant un ensemble (fini) d’étiquettes de classes sémantiques, tels que “personne” et “chaise”. Cependant, pour des instructions ambiguës données à un robot, telles que “apporte-moi une collation saine”, cette approche est insuffisante. En conséquence, des travaux récents ont exploité de grands réseaux de neurones pré-entraînés appelés “modèles de fondation”, dont les représentations latentes apprises offrent plus de flexibilité que les étiquettes de classe, mais ces approches peuvent être inefficaces. Dans ce travail, nous construisons des représentations de scènes 3D qui tirent parti des modèles de fondation pour encoder la sémantique, permettant des requêtes à vocabulaire ouvert et multimodales, tout en restant évolutives et efficaces. Nous présentons initialement ConceptFusion, qui construit des cartes 3D à vocabulaire ouvert en assignant à chaque point 3D un vecteur de caractéristiques qui encode la sémantique, permettant des requêtes nuancées et multimodales, mais à un coût de mémoire élevé. Nous présentons ensuite ConceptGraphs, qui s’appuie sur l’approche précédente avec une structure de graphe de scène qui assigne des vecteurs de caractéristiques sémantiques aux objets au lieu des points, augmentant ainsi l’efficacité, tout en permettant la planification sur le graphe de scène construit. Les deux systèmes ne nécessitent pas d’entraînement supplémentaire ni de réglage fin des modèles, mais permettent aux robots d’effectuer des tâches de recherche et de navigation inédites, comme le montrent nos expériences dans le monde réel. / A central challenge in robotics is building actionable world representations. To perform complex tasks, robots need to build a 3D representation of their environment that represents the geometric, visual, and semantic information of the scene, and is efficient to use. Existing approaches encode semantic information using a (finite) set of semantic class labels, such as “person” and “chair”. However, for ambiguous instructions to a robot, such as “get me a healthy snack”, this approach is insufficient. As a result, recent works have leveraged large pre-trained neural networks called “foundation models”, whose learned latent representations offer more flexibility than class labels, but these approaches can be inefficient. For example, they may require prohibitive amounts of video memory, or an inability to edit the map. In this work, we construct 3D scene representations that leverage foundation models to encode semantics, allowing for open-vocabulary and multimodal queries, while still being scalable and efficient. We initially present ConceptFusion, which builds open-vocabulary 3D maps by assigning each 3D point a feature vector that encodes semantics, enabling nuanced and multimodal queries, but at high memory cost. We then present ConceptGraphs, which builds upon the previous approach with a scene graph structure that assigns semantic feature vectors to objects instead of points, increasing efficiency, while also enabling planning over the constructed scene graph. Both systems do not require any additional training or fine-tuning of models, yet enable novel search and navigation tasks to be performed by robots, as shown by our real world experiments.

Mobile robots need to move in the real world for the majority of tasks. Their control is often intertwined with the tasks they have to solve. Unforeseen events must have an adequate and prompt reaction, in order to solve the corresponding task satisfactorily. A robust system must be able to respond to a variety of events with specific solutions and strategies to keep the system running. Robot navigation control systems are essential for this. In this thesis we present a robot navigation control system that fulfills these requirements: Move Base Flex. Furthermore, the map representation used to model the environment is essential for path planning. Depending on the representation of the map, path planners can solve problems like simple 2D indoor navigation, but also complex rough terrain outdoor navigation with multiple levels and varying slopes, if the corresponding representation can model them accurately. With Move Base Flex, we present a middle layer navigation framework for navigation control, that is map independent at its core. Based on this, we present the Mesh Navigation Stack to master path planning in complex outdoor environments using a developed mesh map to model surfaces in 3D. Finally, to solve path planning in complex outdoor environments, we have developed and integrated the Continuous Vector Field Planner with the aforementioned solutions and evaluated it on five challenging and complex outdoor datasets in simulation and in the real-world. Beyond that, the corresponding developed software packages are open source available and have been released to easily reproduce the provided scientific results.