Automated driving can help solve the current and future problems of individualtransportation. Automated valet parking is a possible approach to help with overcrowded parking areas in cities and make electric vehicles more appealing. In an automated valet system, drivers are able to drop off their vehicle close to a parking area. The vehicle drives to a free parking spot on its own, while the driver is free to perform other tasks — such as switching the mode of transportation. Such a system requires the automated car to navigate unstructured, possibly three dimensional areas. This goes beyond the scope ofthe tasks performed in the state of the art for automated driving. This thesis describes a visual localization system that provides accuratemetric pose estimates. As sensors, the described system uses multiple monocular cameras and wheel-tick odometry. This is a sensor set-up that is close to what can be found in current production cars. Metric pose estimates with errors in the order of tens of centimeters enable maneuvers such as parking into tight parking spots. This system forms the basis for automated navigationin the EU-funded V-Charge project. Furthermore, we present an approach to the challenging problem of life-long mapping and localization. Over long time spans, the visual appearance ofthe world is subject to change due to natural and man-made phenomena. The effective long-term usage of visual maps requires the ability to adapt to these changes. We describe a multi-session mapping system, that fuses datasets intoiiia single, unambiguous, metric representation. This enables automated navigation in the presence of environmental change. To handle the growing complexityof such a system we propose the concept of Summary Maps, which contain a reduced set of landmarks that has been selected through a combination of scoring and sampling criteria. We show that a Summary Map with bounded complexity can achieve accurate localization under a wide variety of conditions. Finally, as a foundation for lifelong mapping, we propose a relational database system. This system is based on use-cases that are not only concerned with solving the basic mapping problem, but also with providing users with a better understanding of the long-term processes that comprise a map. We demonstrate that we can pose interesting queries to the database, that help us gain a better intuition about the correctness and robustness of the created maps. This is accomplished by answering questions about the appearance and distribution of visual landmarks that were used during mapping. This thesis takes on one of the major unsolved challenges in vision-based localization and mapping: long-term operation in a changing environment. We approach this problem through extensive real world experimentation, as well as in-depth evaluation and analysis of recorded data. We demonstrate that accurate metric localization is feasible both during short term changes, as exemplified by the transition between day and night, as well as longer term changes, such as due to seasonal variation.

Submitted by Fernando Souza (fernandoafsou@gmail.com) on 2017-08-10T11:47:56Z No. of bitstreams: 1 arquivototal.pdf: 11498612 bytes, checksum: 9182e9402b0905c4209bc405c726f8cc (MD5) / Made available in DSpace on 2017-08-10T11:47:56Z (GMT). No. of bitstreams: 1 arquivototal.pdf: 11498612 bytes, checksum: 9182e9402b0905c4209bc405c726f8cc (MD5) Previous issue date: 2017-02-17 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / With the advance in the research of mobile robots localization algorithms, the need for natural landmark identification and recognition has increased. The detection of natural landmarks is a challenging task because their appearance can be different in shape and design and, as well, they suffer influence of the environment illumination. As an example, a typical 2D object recognition algorithm may not be able to handle the large optical variety of doors and staircases in large corridors. On another direction, recent improvements in low-cost 3D sensors (of the type RGB-D) enable robots to perceive the environment as a 3D spatial structure. Thus, using this new technology, an algorithm for natural landmark identification and recognition based on images acquired from an RGB-D camera is proposed. Basically, during the identification phase that is a first step for working with landmarks, the algorithm exploits the basic structural knowledge about the landmarks by extracting their edges and creating a cloud of edge points. In the next, the recognition phase, the edges are used with a proposed on-the-fly unsupervised recognition algorithm in order to demonstrate the effectiveness of the approach in recognizing doors and staircases. Two methods of recognition have been proposed and results show that a general technique of the two methods passes from the 96 of accuracy. Future approaches propose a mix of these two methods for better results of recognition, as well as inclusion of new objects such as drinking fountains, dumps and compare this modified approach with other approaches that require training, such as nearest K-neighbors, Bayes and neural networks . / Com o avanço na pesquisa de algoritmos de localização de robôs móveis, a necessidade de identificação e reconhecimento de pontos de referência naturais aumentou. A detecção de marcos naturais é uma tarefa desafiadora, porque a sua aparência pode ser diferente em forma e design e, também, eles sofrem influência da iluminação do ambiente. Como um exemplo, um algoritmo de reconhecimento de objeto 2D típico pode não ser capaz de lidar com a grande variedade óptica de portas e escadas em corredores grandes. Em outra direção, as melhorias recentes em sensores 3D de baixo custo (do tipo RGB-D) permitem aos robôs perceber o ambiente como uma estrutura espacial 3D. Assim, usando esta nova tecnologia, um algoritmo para identificação e reconhecimento de marco natural baseado em imagens adquiridas a partir de uma câmera RGB-D é proposto. Basicamente, durante a fase de identificação que é um primeiro passo para trabalhar com marcos, o algoritmo explora o conhecimento estrutural básico sobre os pontos de referência, extraindo suas bordas e criando uma nuvem de pontos de borda. No próxima, a fase de reconhecimento, as arestas são usadas com um algoritmo de reconhecimento não supervisionado proposto on-the-fly para demonstrar a eficácia da abordagem no reconhecimento de portas e escadarias. Dois métodos de Reconhecimento foram propostos e resultados mostram que a eficiência geral dos dois métodos passa dos 96% de Precisão de reconhecimento. Abordagens futuras propõem-se a fusão dos dois métodos para melhores resultados no reconhecimento, bem como inclusão de novos objetos como bebedouros, lixeiras e comparar essa abordagem modificada com outras abordagens que necessitam de treinamento, como K-Neighbouring mais próximo, Bayes e redes neurais.

L'objectif de cette thèse est de réaliser des missions diverses de navigation autonome en environnement intérieur et encombré avec des robots terrestres. La perception de l'environnement est assurée par un banc stéréo embarqué sur le robot et permet entre autres de calculer la localisation de l'engin grâce à un algorithme d'odométrie visuelle. Mais quand la qualité de la scène perçue par les caméras est faible, la localisation visuelle ne peut pas être calculée de façon précise. Deux solutions sont proposées pour remédier à ce problème. La première solution est l'utilisation d'une méthode de fusion de données multi-capteurs pour obtenir un calcul robuste de la localisation. La deuxième solution est la prédiction de la qualité de scène future afin d'adapter la trajectoire du robot pour s'assurer que la localisation reste précise. Dans les deux cas, la boucle de commande est basée sur l'utilisation de la commande prédictive afin de prendre en compte les différents objectifs de la mission : ralliement de points, exploration, évitement d'obstacles. Une deuxième problématique étudiée est la navigation par points de passage avec évitement d'obstacles mobiles à partir des informations visuelles uniquement. Les obstacles mobiles sont détectés dans les images puis leur position et vitesse sont estimées afin de prédire leur trajectoire future et ainsi de pouvoir anticiper leur déplacement dans la stratégie de commande. De nombreuses expériences ont été réalisées en situation réelle et ont permis de montrer l'efficacité des solutions proposées. / The aim of this thesis is to perform various autonomous navigation missions in indoor and cluttered environments with mobile robots. The environment perception is ensured by an embedded stereo-rig and a visual odometry algorithm which computes the localization of the robot. However, when the quality of the scene perceived by the cameras is poor, the visual localization cannot be computed with a high precision. Two solutions are proposed to tackle this problem. The first one is the data fusion from multiple sensors to perform a robust computation of the localization. The second solution is the prediction of the future scene quality in order to adapt the robot's trajectory to ensure that the localization remains accurate. In the two cases, the control loop is based on model predictive control, which offers the possibility to consider simultaneously the different objectives of the mission : waypoint navigation, exploration, obstacle avoidance. A second issue studied is waypoint navigation with avoidance of mobile obstacles using only the visual information. The mobile obstacles are detected in the images and their position and velocity are estimated in order to predict their future trajectory and consider it in the control strategy. Numerous experiments were carried out and demonstrated the effectiveness of the proposed solutions.