Dans un travail réalisé au Centre de Robotique et à l'Institut VEDECOM, nous nous sommes intéressés aux systèmes robustes de localisation visuelle en milieu urbain pour la voiture autonome. Obtenir une pose exacte à partir d'une mono-caméra est difficile et insuffisant en terme de précision pour la voiture autonome actuelle. Nous nous sommes concentrés sur l'utilisation de Systèmes d'Information Géographiques (SIG) pour concevoir une approche fiable, précise et absolue de localisation en milieu urbain.Le développement de SIG publics nous a apporté un nouvel horizon pour résoudre le problème de la localisation, mais ses informations, telles que les cartes topologiques, sémantiques, métriques, les Street Views, les cartes de profondeur, les cartes cadastrales 3D et les cartes en haute définition, doivent être bien analysées et organisées pour extraire les informations pertinentes pour une voiture autonome. Notre première tâche consistait à concevoir une base de données hors ligne accessible par un robot à partir d'un SIG public dense, à savoir Google Maps, qui a l'avantage d'avoir une couverture mondiale. Nous générons une représentation topométrique compacte de l'environnement urbain en extrayant quatre données utiles du SIG, y compris : les topologies, les géo-coordonnées, les Street Views panoramiques et les cartes de profondeur associées. Dans le même temps, un ensemble de données en ligne a été acquis par une mono-caméra équipée sur les véhicules de VEDECOM. Afin de rendre les Street View sphériques compatibles avec l'imagerie en ligne, une transformation basée sur l'interpolation d'image est introduite pour obtenir des images rectilignes à partir de Street Views.Nous proposons deux méthodes de localisation : l'une est une approche de vision par ordinateur basée sur l'extraction de caractéristiques, l'autre est une méthode d'apprentissage basée sur les réseaux de neurones convolutionnels (convnet). En vision par ordinateur, l'extraction de caractéristiques est un moyen populaire de résoudre le positionnement à partir d'images. Nous tirons parti de Google Maps et utilisons ses données topo-métriques hors ligne pour construire un positionnement grossier à fin, à savoir un processus de reconnaissance de lieu topologique puis une estimation métrique de pose par optimisation de graphe. La méthode a été testée en environnement urbain et démontre à la fois une précision sous-métrique et une robustesse aux changements de point de vue, à l'illumination et à l'occlusion. Aussi, les résultats montrent que les emplacements éloignés de Street Views produisent une erreur significative dans la phase d'estimation métrique. Ainsi, nous proposons de synthétiser des Street Views artificielles pour compenser la densité des Street View originales et améliorer la précision.Cette méthode souffre malheureusement d'un temps de calcul important. Étant donné que le SIG nous offre une base de données géolocalisée à l'échelle mondiale, cela nous motive à régresser des localisations globales directement à partir d'un convnet de bout en bout. La base de données hors ligne précédemment construite est encore insuffisante pour l'apprentissage d'un convnet. Pour compenser cela nous densifions la base d'origine d'un facteur mille et utilisons la méthode d'apprentissage par transfert pour faire converger notre régresseur convnet et avoir une bonne performance. Le régresseur permet également d'obtenir une localisation globale à partir d'une seule image et en temps réel.Les résultats obtenus par ces deux approches nous fournissent des informations sur la comparaison et la relation entre les méthodes basées sur des caractéristiques et celles basées sur le convnet. Après avoir analysé et comparé les performances de localisation des deux méthodes, nous avons également abordé des perspectives pour améliorer la robustesse et la précision de la localisation face au problème de localisation urbaine assistée par SIG. / In a work made at Centre de Robotique and Institut VEDECOM, we studied robust visual urban localization systems for self-driving cars. Obtaining an exact pose from a monocular camera is difficult and cannot be applied to the current autonomous cars. We mainly focused on fully leveraging Geographical Information Systems (GIS) to achieve a low-cost, robust, accurate and global urban localization.The development of public GIS's has brought us a new horizon to address the localization problem but their tremendous amount of information, such as topological, semantic, metric maps, Street Views, depth maps, 3D cadastral maps and High Definition maps, has to be well analyzed and organized to extract relevant information for self-driving cars. Our first task was to design a robotic accessible offline database from a dense public GIS, namely Google Maps, which has the advantage to propose a worldwide coverage. We make a compact topometric representation for the dynamic urban environment by extracting four useful data from the GIS, including topologies, geo-coordinates, panoramic Street Views, and associated depth maps. At the same time, an online dataset was acquired with a low-cost camera equipped on VEDECOM vehicles. In order to make spheric Street Views compatible with the online imagery, an image warping and interpolation based transformation is introduced to render rectilinear images from Street Views.We proposed two localization methods: one is a handcrafted-features-based computer vision approach, the other is a convolutional neural network (convnet) based learning technique. In computer vision, extracting handcrafted features is a popular way to solve the image based positioning. We take advantages of the abundant sources from Google Maps and benefit from the topometric offline data structure to build a coarse-to-fine positioning, namely a topological place recognition process and then a metric pose estimation by a graph optimization. The method is tested on an urban environment and demonstrates both sub-meter accuracy and robustness to viewpoint changes, illumination and occlusion. Moreover, we demonstrate that sparse Street View locations produce a significant error in the metric pose estimation phase. Thus our former framework is refined by synthesizing more artificial Street Views to compensate the sparsity of original Street Views and improve the precision.The handcrafted feature based framework requires the image retrieval and graph optimization. It is hard to achieve in a real-time application. Since the GIS offers us a global scale geotagged database, it motivates us to regress global localizations from convnet features in an end-to-end manner. The previously constructed offline database is still insufficient for a convnet training. We hereby augment the originally constructed database by a thousand factor and take advantage of the transfer learning method to make our convnet regressor converge and have a good performance. In our test, the regressor can also give a global localization of an input camera image in real time.The results obtained by the two approaches provide us insights on the comparison and connection between handcrafted feature-based and convnet based methods. After analyzing and comparing the localization performances of both methods, we also talked about some perspectives to improve the localization robustness and precision towards the GIS-aided urban localization problem.

Reyas手工珠寶：從生產到品牌經營 / Reyas Handcrafted Jewelry is the future of ‘Bead House’. Currently, Bead House (manufacturer of handcrafted jewelry) functions as a wholesaler selling to various regions such as – Morocco, Thailand, India, Italy, Brazil and USA. The export of handcrafted jewelry is growing in Nepal with more buyers coming in every month that want to purchase both traditional designs in countries such as India and Morocco and contemporary designs in countries like Italy and USA. Demand is always increasing but due to limited manufacturing capability and marketing capability the company sees the need to expand. In 2015, Bead House will expand its manufacturing capability delivering double the output thereby increasing sales in volume. Currently, most of the additional demand is being met through purchasing other competitors inventory. Additionally, Bead House will rebrand as Reyas Handcrafted Jewelry (RHJ) developing more innovative and creative designs which will be sold through our partnership with Hotels in Nepal and also through our website. These designs will not be available in bulk and will be sold off as unique individual pieces. This business plan looks at setting up additional manufacturing capability, hiring designers to develop and craft beautiful and modern handcrafted jewelry, marketing the brand RHJ through social media, its website and distribution network of hotels. Target customers are wholesalers who buy in bulk and individuals who would like to buy the most unique product or jewelry accessory available in the market. RHJ plans to be the lead exporter of handcrafted jewelry within the next five years.

A number of researchers have shown that ear recognition is a viable alternative to more common biometrics such as fingerprint, face and iris because the ear is relatively stable over time, the ear is non-invasive to capture, the ear is expressionless, and both the ear’s geometry and shape have significant variation among individuals. Researchers have used different approaches to enhance ear recognition. Some researchers have improved upon existing algorithms, some have developed algorithms from scratch to assist with recognizing individuals by ears, and some researchers have taken algorithms tried and tested for another purpose, i.e., face recognition, and applied them to ear recognition. These approaches have resulted in a number of state-of-the-art effective methods to identify individuals by ears. However, most ear recognition research has been done using ear images that were captured in an ideal setting: ear images have near perfect lighting for image quality, ears are in the same position for each subject, and ears are without earrings, hair occlusions, or anything else that could block viewing of the entire ear. In order for ear recognition to be practical, current approaches must be improved. Ear recognition must move beyond ideal settings and demonstrate effectiveness in an unconstrained environment reflective of real world conditions. Ear recognition approaches must be scalable to handle large groups of people. And, ear recognition should demonstrate effectiveness across a diverse population. This dissertation advances ear recognition from ideal settings to real world settings. We devised an ear recognition framework that outperformed state-of-the-art recognition approaches using the most challenging sets of publicly available ear images and the most voluminous set of unconstrained ear images that we are aware of. We developed a Convolutional Neural Network-based solution for ear normalization and description, we designed a two-stage landmark detector, and we fused learned and handcrafted descriptors. Using our framework, we identified some individuals that are wearing earrings and that have other occlusions, such as hair. The results suggest that our framework can be a gateway for identification of individuals in real world conditions.

The semi-arid region of the Northeast of Brazil is a place known for its climactic and social difficulties since the colonization period and its main activity is the production of cheese ( queijo coalho and queijo manteiga ). This work is the result of the changes on the structure of the handcrafted cheese factories, especially because of the demanding of the rules concerning hygiene and food safety on dairy products. Its main goal is to identify the typological categories and relate them to the way the handcrafted cheese is made ( queijo coalho and queijo manteiga ), according to the regulations. The rural architecture was based on the connection between the environment built, the productive activity and the culture, in an effort to identify the constructive specificities of the local level. The model called co-evolução was used as the theoretical base. The field research was done in eleven rural cheese properties in the town of Major Izidoro, Alagoas, using physical and photographic data, subsidizing the investigation of productive and spatial. The criteria used were based on the rules of the (BPF). The architectonic cheese typologies were classified in three groups: the traditional (handcrafted cheese), the intermediate (handcrafted cheese or dairy) and the industrial (dairy), in relation to the inside and outside of the property, its composition and its organization. When it comes to the productive aspect, it was taken into consideration that the types of cheese are due to different processes in fabrication. The multidisciplinary theme suggests researches in social areas and in architecture, and it may explore more cheese factories in the state of Alagoas as in other regions of the Northeast. / A região semi-árida do nordeste brasileiro é uma região marcada pelas dificuldades climática e social desde o período da colonização e tem como atividade produtiva tradicional a fabricação dos queijos de coalho e de manteiga. Este trabalho surgiu das indagações a respeito das alterações na estrutura física das queijarias artesanais, decorrentes principalmente, das exigências às normas que tratam da higiene e segurança alimentar sobre os produtos lácteos. Tem como objetivo identificar as categorias tipológicas e relacioná-las ao modo de produção das queijarias artesanais de queijo de coalho e de queijo de manteiga, verificando o atendimento as regulamentações. A caracterização da arquitetura rural fundamentou-se na conexão entre o ambiente construído, a atividade produtiva e a cultura, no esforço de identificarem-se as especificidades construtivas ao nível local. Teve como base teórica, o modelo denominado de co-evolução. A pesquisa de campo ocorreu em onze propriedades rurais queijeiras do município de Major Izidoro, Alagoas, mediante levantamento físico e fotográfico, subsidiando a investigação dos vetores produtivo e espacial. Utilizaram-se como critérios os itens da norma (BPF). Classificou-se as tipologias arquitetônicas das queijarias em três grupos: o tradicional (fabriqueta), intermediário (fabriqueta ou laticínio) e industrial (laticínio), com respeito à parte interna e externa da propriedade, composição e organização. Quanto ao aspecto produtivo, considerou-se que os tipos de queijo são decorrentes dos diferentes processos de fabricação. O tema multidisciplinar sugere pesquisas nas áreas sociais e na arquitetura, podendo explorar mais queijarias tanto no estado de Alagoas, quanto em outras regiões do Nordeste.