The population growth associated with natural resources shortage, food demand increasing and the lack of manpower for agricultural activities generate important challenges to agriculture and engineering. The challenges imply the improvement of productivity with fewer resources, This scenario is consisted parameters that generally are in opposite directions. This work presents the development of a rover to agriculture (R2A) to support agriculturescouting tasks, the tasks will be in the future important tools to improve the productivity and enable the use of less agricultural supplies. The research presents the development of a rover for agriculture (R2A) to support scouting tasks; tasks that in the future will improve productivity and allowed the use of less agricultural supplies. The study begins with a bibliographic review of Robots for agriculture, Rovers, and agricultural Rovers developed by different research institutions. After the review is presented the work methodology based on mechanic and mechatronic design methodologies; in the development of the methodology are presented the general crop characteristics, the proposed of scouting tasks, the benchmarking developing mathematical models, CAD (Computer Aided Design ) models, simulations and tests in order to know the different features of the Rovers and agricultural robots. Using the knowledge gained in the course of work is proposed the concept of a rover for agriculture R2A, the concept is compared in simulations, and developed the detailed design using CAE tools (Computer Aided Engineering) after it built a prototype and tested. As results are presented comparative simulations of R2A, their mathematical modeling, R2A simulations in ideally conditions highlighting their skills, and finally R2A tests and comparison are presented. / O crescimento populacional associado à escassez de recursos naturais, a crescente demanda alimentar e a falta de mão de obra para as actividades agrícolas geram importantes desafios para a agricultura e a engenharia. Os desafios implicam a melhoria da produtividade com menos recursos. O cenário é constituido por parâmetros que geralmente estão em direções opostas. O trabalho apresenta o desenvolvimento de um rover para agricultura (R2A) para suportar tarefas de Scouting, tarefas que no futuro melhorarão a produtividade e permitirão o uso de menos subministros agrícolas. O estudo começa pela revisão bibliográfica de Robôs para agricultura, Rovers e Rovers agrícolas desenvolvidos por diferentes instituições de pesquisa. Apos a revisão é apresentada a metodologia do trabalho baseada nas metodologias de projeto mecânico e mecatrônico; no desenvolvimento da metodologia são apresentadas as caraterísticas das culturas de maneira geral, a proposta de tarefas de Scouting, o benchmarking desenvolvendo modelos matemáticos, modelos CAD (Computer Aided Design) simulações e testes com o intuito de conhecer as diferentes caraterísticas dos Rovers e Robôs agrícolas. Usando o conhecimento no decorrer do trabalho é proposto o conceito do rover para agricultura R2A, o conceito é comparado em simulações, e feito o projeto detalhado usando ferramentas CAE (Computer Aided Enginnerring), após é construído o protótipo, e testado. Como resultados são apresentadas simulações comparativas do R2A, a sua modelagem matemática, simulações do R2A em condições ideais, destacando as suas capacidades, e finalmente são apresentados os testes e comparações do R2A.

Este trabalho aborda o problema de planejamento de caminhos em robótica móvel autônoma utilizando campos potenciais. Dentre as várias técnicas de campos potenciais para controlar robôs, encontram-se as técnica de Campos Potenciais de Khatib1 (CP), Campo Potencial Harmônico (CPH), Campo Potencial Orientado (CPO) e Campo Potencial Localmente Orientado (CPLO). As técnicas CPH, CPO e CPLO são chamadas de técnicas baseadas em Problema de Valor de Contorno (PVC), pois são obtidas a partir de soluções de Equações Diferenciais Parciais (EDP) Elípticas em uma determinada condição de contorno, é obtido um sistema planejador de caminhos. Tais técnicas necessitam de uma etapa de solução de sistemas lineares, na qual se utiliza métodos iterativos, decorrentes da aplicação do método de diferenças finitas como solucionador das EDP. No presente trabalho, as técnicas de Campos Potenciais baseados em PVC foram estudadas e implementadas (usando processamento sequencial e paralelo), de modo a obter resultados de forma mais rápida e confiável. Foram utilizadas arquiteturas paralelas do tipo manycore. Finalmente, são feitas análises comparativas entre os vários métodos implementados. Todos os métodos estão prontos para serem incorporados tanto no simulador quanto nos times de robôs em desenvolvimento pelo grupo Warthog Robotics / This works details the task o path planning in autonomous mobile robots using potential fields techniques. Among potential fields techniques to control robots, there are Khatibs Potential Field2 (KPF), Harmonic Potential Field (HPF), Oriented Potential Field (OPF) and Locally Oriented Potential Field (LOPF). The HPF, OPF and LOPF techniques are called Boundary Value Problem (BVP) based, bacause they are obtained from numerical solutions of Elliptic Partial Differential Equations (PDE) in a well-defined boundary condition. These techniques go through a step of solving linear systems, in which is used iterative methods, that came from numerical solution of PDE. In this work, potential fields BVP based was studied and coded (using sequential and parallel architectures), to obtain results more quickly and reliably. And, finally, a comparative analyses of the various methods implemented are made. All methods are ready to be incorporated in the intelligent systens that are being developed by Warthog Robotics

La performance est un concept multiforme largement décliné et défini en robotique de manipulation où il relève de contraintes spécifiques : environnement non dynamique de dimensions limitées, énergie infinie. Les indicateurs de performance proposés dans le cadre de missions robotiques terrestres sont moins largement acceptés.Pour notre part nous les déclinons en axes principaux : énergie, sécurité, localisation, stabilité, et secondaires : durée par exemple.Dans le cadre de ce travail nous nous intéressons à des missions robotiques soumises à des contraintes de performance liées à la durée, la sécurité et l'énergie dans un environnement connu mais dynamique.La déclinaison des contraintes imposées à la mission et au robot permet de décomposer la mission en une suite d'activités aux contraintes invariantes. Chacune de ces activités pouvant être réalisée à l'aide d'un ensemble de tâches robotiques (se déplacer, se localiser, analyser une image, etc.) pouvant elles-mêmes être implémentées de différentes façons en fonction des actionneurs, capteurs ou algorithmes pouvant être utilisés.Le problème adressé est le suivant : comment choisir les ressources matérielles et logicielles à utiliser tout au long de la mission de façon à vérifier les contraintes de performance imposées ? C'est un problème de sac à dos multicritères NP-Complet et l'espace de recherche devient très rapidement inexplorable. De façon à proposer et garantir une solution applicable sous des contraintes temps réel, nous faisons appel à un algorithme permettant de trouver un ensemble ordonné de bonnes solutions en temps linéaire.L'approche de gestion de ressources proposée a été implémentée sur un robot Pioneer-3DX et une architecture de contrôle s'appuyant sur le Middleware ContrACT.Cette approche a été validée dans le cadre d'une mission de patrouille d'une longueur de 200 m et durant une dizaine de minutes, au sein du laboratoire LIRMM, pour vérifier l'état de vannes.Pour la mission choisie, l'espace d'états à considérer est supérieur à 10^{14}.Tout au long de la mission, les ressources matérielles et logicielles sont choisies dynamiquement et d'une façon autonome afin de satisfaire les contraintes de performance imposées.Si des ressources ne sont plus opérationnelles, ou si trop d'évitements d'obstacles non prévus sont effectués, l'approche développée est capable de trouver en ligne une nouvelle solution d'affectation des ressources vérifiant les contraintes de performance imposées, si elle existe.Ces travaux participant donc à l'accroissement du niveau de tolérance aux fautes du système robotisé. / The performance is a multi-form concept widely defined in manufacturing robotics with specific environment conditions (static and perfectly known) and infinite energy).However, performance indicators proposed in mobile robotics are less widely accepted.We differenciate between main performance axes (energy, safety, localization and stability) and secondary performance axes (duration for example).In our work, we are interested on missions realized under duration, safety and energy performance constraints, in a known but dynamic environment.Applying the different constraints decomposes the mission into a sequence of activities realized under invariant constraints.Each one, can be realized by a set of robotic tasks (move, be located, analyze an image, etc.).These tasks can be implemented in various ways according to the different possibles actuators, sensors and algorithms configurations.The adressed problem is the following: how to choose the hardware and software resources to use along a mission while satisfying the different performance constraints ? It is a multicriteria knapsack problem known to be NP-hard, where the complexity becomes very quickly unexplorable.To propose and guarantee an applicable solution under real-time constraints, we used an algorithm allowing to find a set of good solutions in few iterations.The proposed resources management approach is implemented on a Pioneer-3DX robot using a control architecture based on the Middleware ContrACT.This approach has been validated on a patrolling mission travelling 200 m within the LIRMM laboratory during about 10 mn, to verify the state of valves.For the considered mission, the state space dimension is higher than 10^{14}.The hardware and software resources are dynamically and autonomously selected along the mission to satisfy the different performance constraints.If a resource becomes faulty and/or many obstacle avoidances occure and lead to performance drift, the developed approach finds on line a new resources allocation solution (if it exists).So this approach allows to enhance the fault tolerance of the robotic system.

Apesar da popularidade das câmeras onidirecionais aplicadas à robótica móvel e da importância do reconhecimento de objetos no universo mais amplo da robótica e da visão computacional, é difícil encontrar trabalhos que relacionem ambos na literatura especializada. Este trabalho visa avaliar um método para classificação de objetos em imagens onidirecionais, analisando sua eficácia e eficiência para ser aplicado em tarefas de auto-localização e mapeamento de ambientes feitas por robôs moveis. Tal método é construído a partir de um classificador de objetos, implementado através de máquinas de suporte vetorial, estendidas para a utilização de Aprendizagem de Múltiplos Núcleos. Também na construção deste método, uma etapa de retificação é aplicada às imagens onidirecionais, de modo a aproximá-las das imagens convencionais, às quais o classificador utilizado já demonstrou bons resultados. A abordagem de Múltiplos Núcleos se faz necessária para possibilitar a aplicação de três tipos distintos de detectores de características em imagens, ponderando, para cada classe, a importância de cada uma das características em sua descrição. Resultados experimentais atestam a viabilidade de tal proposta. / Despite the popularity of omnidirectional cameras used in mobile robotics, and the importance of object recognition in the broader universe of robotics and computer vision, it is difficult to find works that relate both in the literature. This work aims at performing the evaluation of a method for object classification in omnidirectional images, evaluating its effectiveness and efficience considering its application to tasks of self-localization and environment mapping made by mobile robots. The method is based on a multiple kernel learning extended support vector machine object classifier. Furthermore, an unwrapping step is applied to omnidirectional images, to make them similar to perspective images, to which the classifier used has already shown good results. The Multiple Kernels approach is necessary to allow the use of three distinct types of feature detectors in omnidirectional images by considering, for each class, the importance of each feature in the description. Experimental results demonstrate the feasibility of such a proposal.

A detecção e reconhecimento de objetos é uma tarefa fundamental em aplicações relacionadas à navegação autônoma de robôs móveis e veículos inteligentes. Com a evolução tecnológica nos sistemas sensoriais, surgiram equipamentos capazes de detectar e representar os elementos presentes no ambiente de forma tridimensional, em estruturas chamadas nuvem de pontos. Os sensores 3D geralmente capturam um grande volume de pontos em curtos intervalos de tempo, o que demanda técnicas robustas para processamento dessa informação além de tolerância a eventuais ruídos nos dados. Uma abordagem frequentemente utilizada na área de Visão Computacional para redução de dimensionalidade é a extração de features robustas, armazenando um subconjunto de informações representativas e simplificadas do conjunto de dados. Esta tese apresenta uma metodologia de classificação de objetos em nuvens de pontos 3D através da extração de features 3D globais. Foi desenvolvido um novo descritor 3D invariante à escala, translação e rotação denominado 3D-CSD (3D-Contour Sample Distances) para representação da superfície dos objetos presentes no ambiente, e utilizado um método de aprendizado supervisionado para reconhecimento de padrões. Os experimentos realizados envolveram o uso de Redes Neurais Artificiais para o reconhecimento de diferentes classes de objetos, avaliando e validando a metodologia proposta. Os resultados obtidos demostraram a viabilidade da aplicação desta abordagem para o reconhecimento de objetos em sistemas de percepção 3D. / Objects detection and recognition is a critical task in applications for mobile robots and intelligent vehicles autonomous navigation. With the advent of many 3D sensors, environment elements can be detected and represented in three-dimensional mode, in structures known as point clouds. 3D sensors usually capture a large amount of points at high rates, requiring robust techniques to process this information and also deal with noise on input data. A common approach in the Computer Vision field for dimensionality reduction is the use of robust features extraction techniques. This way, only a subset with representative and simplified information from the dataset is processed. This thesis presents a methodology for objects recognition in 3D point clouds using global 3D features extraction. A novel 3D descriptor invariant to scale, translation and rotation named 3D-CSD (3D-Contour Sample Distances) was developed to represent the objects surface, and a supervised learning method used for pattern recognition. The experiments were performed using Artificial Neural Networks for the recognition of different classes of objects, evaluating and validating the proposed methodology. Obtained results demonstrated the feasibility of this approach application for object recognition in 3D perception systems.

De acordo com a Organização Mundial da Saúde,cerca de 1,2milhões de pessoas no mundo morrem em acidentes de trânsito. Sistemas de assistência ao motorista e veículos autônomos podem diminuir o número de acidentes. Dentre as várias demandas existentes para viabilizar essa tecnologia, sistemas computacionais de percepção ainda permanecem sem uma solução definitiva. Dois deles, detecção de obstáculos e de via navegável, normalmente fazem uso de algoritmos sofisticados como técnicas de aprendizado supervisionado, que mostram resultados impressionantes quando treinados com bases de dados bem definidas e diversificadas.Entretanto, construir, manter e atualizar uma base de dados com exemplos de vários lugares do mundo e em diversas situações é trabalhoso e complexo. Assim, métodos adaptativos e auto-supervisionados mostram-se como boas alternativas para sistemas de detecção do futuro próximo. Neste contexto, esta tese apresenta um método para estimar obstáculose via navegável através de sensores de baixo custo (câmeras estereoscópicas), sem o uso de técnicas de aprendizado de máquina e de diversas suposições normalmente utilizadas por trabalhos já disponíveis na literatura. Esses métodos utilizando sensor estereoscópico foram comparados fazendo uso de sensores do tipo 3D-LIDAR e mostraram resultados semelhantes. Este sistema poderá ser usado como uma fase pré-processamento de dados para melhorar ou viabilizar métodos adaptativos de aprendizado. / World wide, an estimated 1.2million lives are lostin road crashes each year and Advanced Driver Assistance Systems (ADAS) and Self-driving cars promise to reduce this number. Among the various issues to complete this technology, perception systems are still an unsolved issues. Normally two of them, obstacle detection and road detection, make use of sophisticated algorithms such as supervised machine learning methods which can perform with impressive results if it was trained with good data sets. Since it is a complex and an expensive job to create and maintain data bases of scenarios from the entire world, adaptive and/or self-supervised methods are good candidates for detection systems in the near future. Due that, this thesis present a method to estimate obsta- cles and estimate the road terrain using low cost sensors (stereo camera), avoiding supervised machine learning techniques and the most common assumptions used by works presented in literature. These methods were compared with 3D-LIDAR approaches achieving similar results and thus it can be used as a pre-processing step to improve or allow adaptive methods with machine learning systems.

Nesta pesquisa são estudados aspectos relacionados à problemática da localização robótica, e um sistema de localização robótica é construído. Para determinação da localização de um robô móvel em relação a um mapa topológico do ambiente, é proposta uma solução determinística. Esta solução é empregada a fim de prover localização para problemas de rastreamento de posição, embora seja de interesse também a observação da eficácia, do método proposto, frente a problemas de localização global. O sistema proposto baseia-se no uso de vetores de atributos, compostos de medições momentâneas extraídas do ambiente através de sensoriamentos pertencentes à percepção do robô. Estimativas feitas a partir da odometria e leitura de sensores de ultra-som são utilizadas em conjunto nestes vetores de atributos, de forma a caracterizar as observações feitas pelo robô. Uma bússola magnética também é empregada na solução. O problema de localização é então resolvido como um problema de reconhecimento de padrões. A topologia do ambiente é conhecida, e a correlação entre cada local neste ambiente e seus atributos são armazenados através do uso de redes neurais artificiais. O sistema de localização foi avaliado de maneira experimental, em campo, em uma plataforma robótica real, e resultados promissores foram obtidos e são apresentados. / In this research aspects related to the robot localization problem have been studied. In order to determine the localization of a mobile robot in relation to a topological map of its environment, a deterministic solution has been proposed. This solution is applied to provide localization for position tracking problems, although it is also of interest to observe the performance of the proposed method applied to global localization problems. The proposed system is based on feature vectors, which are composed of momentaneous measures extracted from sensory data of the robots perception. Estimative made from odometry, sonars and magnetic compass readings are used together in these feature vectors, in order to characterize observed scenes by the robot. Thus, the localization problem is solved as a pattern recognition problem. The topology of the environment is known, and the correlation between each place of this environment and its features is stored using an artificial neural network. The localization system was experimentally evaluated, in a real robotic platform. The results obtained allow validation of the methodology.

In order to enable more widespread application of robots, we are required to reduce the human effort for the introduction of existing robotic platforms to new environments and tasks. In this thesis, we identify three complementary strategies to address this challenge, via the use of imitation learning, domain adaptation, and transfer learning based on simulations. The overall work strives to reduce the effort of generating training data by employing inexpensively obtainable labels and by transferring information between different domains with deviating underlying properties. Imitation learning enables a straightforward way for untrained personnel to teach robots to perform tasks by providing demonstrations, which represent a comparably inexpensive source of supervision. We develop a scalable approach to identify the preferences underlying demonstration data via the framework of inverse reinforcement learning. The method enables integration of the extracted preferences as cost maps into existing motion planning systems. We further incorporate prior domain knowledge and demonstrate that the approach outperforms the baselines including manually crafted cost functions. In addition to employing low-cost labels from demonstration, we investigate the adaptation of models to domains without available supervisory information. Specifically, the challenge of appearance changes in outdoor robotics such as illumination and weather shifts is addressed using an adversarial domain adaptation approach. A principal advantage of the method over prior work is the straightforwardness of adapting arbitrary, state-of-the-art neural network architectures. Finally, we demonstrate performance benefits of the method for semantic segmentation of drivable terrain. Our last contribution focuses on simulation to real world transfer learning, where the characteristic differences are not only regarding the visual appearance but the underlying system dynamics. Our work aims at parallel training in both systems and mutual guidance via auxiliary alignment rewards to accelerate training for real world systems. The approach is shown to outperform various baselines as well as a unilateral alignment variant.

L'objectif de ma thèse est d'étudier les interactions entre processus cognitifs et émotionnels à travers le prisme de la robotique neuromimétique.Les modèles proposés sont implémentés sur des réseaux de neurones artificiels et encorporés dans des plateformes robotiques -- c'est-à-dire des systèmes encorporés et situés. D'une manière générale, l'intérêt est double : 1) s'inspirer des solutions biologiques pour concevoir des systèmes qui interagissent mieux avec leurs environnements physiques et sociaux, et 2) mettre en place des modèles computationels comme moyen de comprendre la cognition et les émotions biologiques.La première partie du manuscrit aborde la navigation comme cadre permettant d'étudier la cognition biologique et artificielle. Dans le Chapitre 1, je commence par donner un aperçu de la littérature en navigation bio-inspirée. Ensuite, deux problématiques sont traitées. Dans le Chapitre 2, la reconnaissance visuelle de lieux en environnement extérieur est abordée. Dans ce cadre, je propose un modèle basé sur les notions de textit{contextes visuels} et de textit{précédence globale} qui combine des informations locales et holistiques. Puis, dans le Chapitre 3, je considère l'textit{apprentissage interactif} de tâches de navigation à travers une communication homme--robot non-verbale basée sur des signaux visuomoteurs de bas niveau.La deuxième partie du manuscrit s'attaque à la question centrale des interactions entre emotion et cognition. Le Chapitre 4 introduit la recherche sur les émotions comme une entreprise inter-disciplinaire, incluant des théories en psychologie, des découvertes en neurosciences et des modèles computationnels. Dans le Chapitre 5, je propose un textit{modèle conceptuel} des interactions emotion--cognition donc différentes instantiations sont ensuite présentées. Plus précisément, dans le Chapitre 6, je modélise la perception de l'textit{espace péripersonnel} quand elle est modulée par des signaux sensoriels et physiologiques ayant une valence émotionnelle. Enfin, dans le Chapitre 7, j'introduis le concept de textit{métacontrôle émotionnel} comme un exemple d'interaction emotion--cognition. Cela consiste à utiliser des signaux émotionnels élicités par une auto-évaluation pour moduler des processus computationnels -- tels que l'attention ou la sélection de l'action -- dans le but de réguler le comportement.Une idée clé dans cette thèse est que, dans les systèmes autonomes, emotion et cognition ne peuvent pas être séparées. En effet, il est maintenant bien admis que les émotions sont très liées à la cognition, en particulier à travers la modulation de différents processus computationnels. Cela pose la question de savoir si ces modulations se produisent au niveau du traitement sensoriel ou celui de la sélection de l'action. Dans cette thèse, je préconise l'intégration des émotions artificielles dans les architectures robotiques via des influences bidirectionnelles avec les processus sensoriels, attentionnels, moteurs and décisionnels. Ce travail tente de mettre en avant la manière dont cette approche des aspects internes des processus émotionnels peut favoriser une interaction efficace avec l'environnement physique et social du système. / The purpose of my thesis is to study interactions between cognitive and emotional processes through the lens of neuromimetic robotics.The proposed models are implemented on artificial neural networks and embodied in robotic platforms -- that is to say embodied and situated systems. In general, the interest is twofold: 1) taking inspiration from biological solutions to design systems that better interact with their physical and social environments, and 2) providing computational models as a means to better understand biological cognition and emotion.The first part of the dissertation addresses spatial navigation as a framework to study biological and artificial cognition. In Chapter 1, I present a brief overview of the literature on biologically inspired navigation. Then, two issues are more specifically tackled. In Chapter 2, visual place recognition is addressed in the case of outdoor navigation. In that matter, I propose a model based on the notions of textit{visual context} and textit{global precedence} combining local and holistic visual information. Then, in Chapter 3, I consider the textit{interactive learning} of navigation tasks through non-verbal human--robot communication based on low-level visuomotor signals.The second part of the dissertation addresses the central question of emotion--cognition interactions. In Chapter 4, I give an overview of the research on emotion as a cross-disciplinary research, including psychological theories, neuroscientific findings and computational models. In Chapter 5, I propose a textit{conceptual model} of emotion--cognition interactions. Then, various instantiations of this model are presented. In Chapter 6, I model the perception of the textit{peripersonal space} when modulated by emotionally valenced sensory and physiological signals. Last, in Chapter 7, I introduce the concept of textit{Emotional Metacontrol} as an example of emotion--cognition interaction. It consists in using emotional signals elicited by self-assessment to modulate computational processes -- such as attention and action selection -- for the purpose of behavior regulation.In this thesis, a key idea is that, in autonomous systems, emotion and cognition cannot be separated. Indeed, it is becoming well admitted that emotion is closely related to cognition, in particular through the modulation of various computational processes. This raises the question of whether this modulation occurs at the level of sensory processing or at the level of action selection. In this thesis, I will advocate the idea that artificial emotion must be integrated in robotic architectures through bidirectional influences with sensory, attentional, decisional and motor processes. This work attempts to highlight how this approach to internal emotional processes could foster efficient interaction with the physical and social environment.

Apesar da popularidade das câmeras onidirecionais aplicadas à robótica móvel e da importância do reconhecimento de objetos no universo mais amplo da robótica e da visão computacional, é difícil encontrar trabalhos que relacionem ambos na literatura especializada. Este trabalho visa avaliar um método para classificação de objetos em imagens onidirecionais, analisando sua eficácia e eficiência para ser aplicado em tarefas de auto-localização e mapeamento de ambientes feitas por robôs moveis. Tal método é construído a partir de um classificador de objetos, implementado através de máquinas de suporte vetorial, estendidas para a utilização de Aprendizagem de Múltiplos Núcleos. Também na construção deste método, uma etapa de retificação é aplicada às imagens onidirecionais, de modo a aproximá-las das imagens convencionais, às quais o classificador utilizado já demonstrou bons resultados. A abordagem de Múltiplos Núcleos se faz necessária para possibilitar a aplicação de três tipos distintos de detectores de características em imagens, ponderando, para cada classe, a importância de cada uma das características em sua descrição. Resultados experimentais atestam a viabilidade de tal proposta. / Despite the popularity of omnidirectional cameras used in mobile robotics, and the importance of object recognition in the broader universe of robotics and computer vision, it is difficult to find works that relate both in the literature. This work aims at performing the evaluation of a method for object classification in omnidirectional images, evaluating its effectiveness and efficience considering its application to tasks of self-localization and environment mapping made by mobile robots. The method is based on a multiple kernel learning extended support vector machine object classifier. Furthermore, an unwrapping step is applied to omnidirectional images, to make them similar to perspective images, to which the classifier used has already shown good results. The Multiple Kernels approach is necessary to allow the use of three distinct types of feature detectors in omnidirectional images by considering, for each class, the importance of each feature in the description. Experimental results demonstrate the feasibility of such a proposal.