Thesis (M.S.)--Rochester Institute of Technology, 2005. / Includes bibliographical references (leaves 98-105).

Orientador: Clesio Luis Tozzi / Tese (doutorado) - Universidade Estadual de Campinas, Faculdade de Engenharia Eletrica e de Computação / Made available in DSpace on 2018-08-10T17:08:02Z (GMT). No. of bitstreams: 1 Nogueira_FernandoMarquesdeAlmeida_D.pdf: 5973806 bytes, checksum: ec2975af264084ab2665aa65eca49541 (MD5) Previous issue date: 2007 / Resumo: Este trabalho apresenta uma nova abordagem iterativa e hierárquica para a obtenção de mapas densos de disparidade, a partir de um par de imagens estereoscópicas retificadas, na qual oclusões são explicitamente identificadas. Esta abordagem se fundamenta num processo de busca hierárquica onde são determinados pares homólogos prioritários sobre os quais é aplicada a Injunção de Ordem. Pares homólogos prioritários consistem nos pontos que apresentam a propriedade de Correspondência Bidirecional para o maior número de janelas de busca organizadas em ordem crescente de tamanho. A aplicação da Injunção de Ordem aos pares homólogos prioritários elimina, já nas primeiras iterações, uma grande quantidade de correspondências ambíguas aumentando o índice de acertos e reduzindo o tempo de processamento. A abordagem proposta destaca-se pela utilização de apenas dois parâmetros que devem ser fornecidos a priori de maneira empírica: o tamanho inicial da janela e o número de níveis hierárquicos. A eficiência do algoritmo proposto foi avaliada a partir de imagens reais e simuladas e os resultados comparados com outras abordagens descritas na literatura / Abstract: This work presents a new hierarchical and iterative approach to obtain dense disparity maps from a pair of rectified stereoscopic images in which occlusions are explicitly identified. This approach is based on a process of hierarchical searching where priority matches are determined and upon which is applied the Order Constraint. Priority matches consist of points that present a property of Bidirectional Correspondence for the greatest number of searching windows, these ones organized in a size ascending order. The Order Constraint application on the priority matches eliminates, usually after the first iterations, a lot of ambiguous matches, increasing the rate of true matches and reducing the execution time. The proposed approach stands out by using only two parameters that must be provided a priori in an empirical way: the initial window size and the number of hierarquical levels. The efficiency of the proposed algorithm was evaluated using real and simulated images and the results were compared to other approaches described in the literature / Doutorado / Automação / Doutor em Engenharia Elétrica

Orientador: Pablo Siqueira Meirelles / Tese (doutorado) - Universidade Estadual de Campinas, Faculdade de Engenharia Mecânica / Made available in DSpace on 2018-08-26T13:45:42Z (GMT). No. of bitstreams: 1 Alvarez-Jacobo_JustoEmilio_D.pdf: 7664743 bytes, checksum: 72e7ab67912a544a9178da02711121d9 (MD5) Previous issue date: 2013 / Resumo: Este trabalho apresenta o estudo e desenvolvimento de um Sistema de Percepção baseado na utilização de sensores telemétricos tipo LIDAR. Uma plataforma de escaneamento a laser em três dimensões LMS-3D é construída a fim da navegação autônoma de robôs. A área navegável é obtida a partir de mapas telemétricos, caracterizados com algoritmos de grades de ocupação (GO) (em duas dimensões com a terceira colorida e 3D) e com o cálculo de gradientes vetoriais. Dois tipos de áreas navegáveis são caracterizadas: (i) área de navegação primária representada por uma área livre dentro da GO; e (ii) área de navegação continua representada pela soma das áreas continuas e gradientes classificados com um determinado limiar. Este limiar indica se uma área é passível de navegação considerando as características do robô. A proposta foi avaliada experimentalmente em ambiente real, contemplou a detecção de obstáculos e a identificação de descontinuidades / Abstract: This thesis was proposed to demonstrate the study and development of a Perception System based on the utilization of a LIDAR telemetric sensors. It was proposed to create a LMS-3D three dimension laser scanning platform, in an attempt to promote the Autonomous Robot Navigation. The scanned area was obtained based on telemetric maps, which was characterized with Occupancy Grid algorithms (OG) (in two dimensions with the third colored and 3D) and Vector Gradients calculation. Two different navigation areas were characterized: (i) primary area of navigation, that represents the free area inside a OG, and (ii) continuous navigation area, that represents the navigated area composed by the sum of continuous areas and the gradients classified by a determined threshold, which indicates the possible navigated area, based on the robot characteristics. The proposition of this thesis was evaluated in a real environment and was able to identify the obstacles detection and also the discontinuance / Doutorado / Mecanica dos Sólidos e Projeto Mecanico / Doutor em Engenharia Mecânica

Master’s thesis is describing main terms in machine vision. It defines the basic working principles, its advantages and disadvantages. It is dividing machine vision systems into several classes. The thesis is solving design of an end effector, of an inspection process and program for smart camera Siemens SIMATIC VS722A, of a program for KUKA KR3 and of mutual communication. At the end the demonstrational application verifying the functionality of entire system is introduced.

Qui n’a jamais souhaité avoir des yeux derrière la tête ? Cette thèse propose d’étudier l’extension du champ visuel humain, que ce soit dans le monde réel ou dans un environnement virtuel. Nous avons d’abord conçu FlyVIZ, un dispositif qui permet d’augmenter le champ visuel. Il est composé d’une caméra catadioptrique, d’un visiocasque et d’un algorithme de traitement d’image. Lorsqu’un utilisateur porte ce dispositif, il dispose d’une vue à 360 degrés de son environnement. Le prototype a été testé avec succès dans différents scenarios, comme attraper un objet tendu dans le dos sans tourner la tête, ou passer des portes en marchant à reculons. Ensuite nous avons proposé une nouvelle méthode pour le rendu d’environnements virtuels, avec un champ visuel étendu, et en temps réel. Pour résoudre les problèmes dus à l’utilisation de projections non planes, nous avons ajouté une nouvelle étape dans le pipeline graphique. Notre méthode a ensuite été adaptée au rendu stéréoscopique avec un champ visuel de 360°. Nous avons mené une évaluation préliminaire sur l’utilisation d’un champ visuel étendu appliqué à une tache de navigation dans des environnements virtuels. Nos résultats semblent confirmer que l’utilisation d’un champ visuel étendu permet une navigation plus efficace, en diminuant le temps moyen pour effectuer une tache. Parmi les différentes projections non planes testées, une préférence pour les projections équirectangulaire et de Hammer a été exprimée. Nous avons également traité le problème de conflit d’indice de profondeurs rencontré dans les images stéréoscopiques lorsqu’un objet affiché en parallaxe négative est partiellement occulté par un bord du support d’affichage. Nous avons proposé SCVC (Stereoscopy Compatible Volume Clipping, le découpage de la scène selon le volume compatible avec la stéréoscopie) pour résoudre ce problème en n’affichant que la partie de l’espace qui n’est pas sujette au conflit d’indice de profondeur. La méthode a été évaluée et les résultats ont montrés que SCVC améliore significativement la perception de la profondeur et que les utilisateurs expriment une préférence pour cette méthode. L’extension du champ visuel humain ouvre de nouvelles perspectives pour l’exploration ou la surveillance de l’environnement d’un utilisateur. Cette extension pourrait bénéficier à diverses applications, que ce soit dans le contexte d’un environnement réel ou virtuel. Pour la sécurité des personnes ou la défense, des pompiers, des policiers ou des soldats pourraient tirer avantage d’un champ visuel étendu. Ce type de visualisation peut également profiter à l’exploration rapide d’environnements virtuels ou à la recherche d’objet dans ces derniers. / Who have never wanted to have eyes in the back of his head? This doctoral thesis proposes to study the extension of the human field-of-view (FoV) in both real and virtual environments. First we have designed FlyVIZ, a new device to increase the human FoV. It is composed of a helmet, combining a catadioptric camera, a HMD and an image processing algorithm. Wearing this device allows a user to experience 360° vision of its surroundings. The prototype is demonstrated through scenarii such as grasping an object held out behind their back without turning their head or walking backward through doorways. Then we have proposed a novel method to render virtual environments with wide FoV in real-time. To solve the rendering issue induced by usage of non-planar projections, we introduce a special stage in real-time rendering pipeline. Our method was then adapted for real-time stereoscopic rendering with 360° FoV. We have conducted a preliminary evaluation of real-time wide FoV rendering for a navigation task in virtual reality. Our results confirm that using a wide FoV rendering method could lead to more efficient navigation in terms of average task completion time. Among the different tested non-planar projection methods, the subjective preference is given to equirectangular and Hammer projections. We also address the problem of frame cancellation, generated by the conflict between two depth cues: stereo disparity and occlusion with the screen border. We have proposed the Stereoscopy Compatible Volume Clipping (SCVC), solving the problem by rendering only the part of the viewing volume free of disparity - frame occlusion conflict. The method was evaluated and results have shown that SCVC notably improved users’ depth perception and that the users expressed preference for SCVC. Wide FoV opens novel perspectives for environments exploration or monitoring. Therefore, it could benefit to several applications, both in real world context or virtual environments. In safety and security applications, firemen, policemen or soldiers could take advantage of wide FoV. Performance of searching task and fast exploration in virtual environments could also be improved with wide FoV.

No description available.

No description available.

Robots with vision capability have been taught to recognize unknown objects by comparing their shape features with those of known objects, which are stored in the vision system as a knowledge base. Traditionally, this knowledge base is created by showing the robot the set of objects that it is likely to come across. This is done with the vision system to be used and must be done in an online mode. An approach to teach the robot in an off-line mode by integrating the robot vision system and an off-line graphic system, has been developed in this research. Instead of showing the objects that the robot is likely to come across, graphic models of the objects were created in an off-line graphic system and a FORTRAN program that processes the models to extract their shape parameters was developed. These shape parameters were passed to the vision system. A program to process an unknown object placed in front of the vision system was developed to extract its shape parameters. A program that compares the parameters of the unknown object with those of the known models was also developed. The vision system was calibrated to measure the pixel dimensions in inches. In the vision system, shape parameters of the objects were found to vary with different orientations. The range of variation for each parameter was established and this was taken into consideration in the parameter comparison program. / Master of Science

M. Tech. Computer System Engineering / Autonomous robots are extensively used equipment in industries and in our daily lives; they assist in manufacturing and production but are used for exploration in dangerous or unknown environments. However for a successful exploration, manufacturing and production, navigation plays an important role. Road detection is a vital factor that assists autonomous robots in perfect navigation. Different methods using camera-vision technique have been developed by various researchers with outstanding results, but their systems are still vulnerable to environmental risks. The frequent weather change in various countries such as South Africa, Nigeria and Zimbabwe where shadow, light intensity and other environmental noises occur on daily basis, can cause autonomous robot to encounter failure in navigation. Therefore, the main research question is: How to enhance the road region detection system to enable an effective and efficient maneuvering of the robot in any weather condition.

In this thesis we investigate visual place categorization by combining successful global image descriptors with a method of visual attention in order to automatically detect meaningful objects for places. The idea behind this is to incorporate information about typical objects for place categorization without the need for tedious labelling of important objects. Instead, the applied attention mechanism is intended to find the objects a human observer would focus first, so that the algorithm can use their discriminative power to conclude the place category. Besides this object-based place categorization approach we employ the Gist and the Centrist descriptor as holistic image descriptors. To access the power of all these descriptors we employ SVM-DAS (discriminative accumulation scheme) for cue integration and furthermore smooth the output trajectory with a delayed Hidden Markov Model. For the classification of the variety of descriptors we present and evaluate several classification methods. Among them is a joint probability modelling approach with two approximations as well as a modified KNN classifier, AdaBoost and SVM. The latter two classifiers are enhanced for multi-class use with a probabilistic computation scheme which treats the individual classifiers as peers and not as a hierarchical sequence. We check and tweak the different descriptors and classifiers in extensive tests mainly with a dataset of six homes. After these experiments we extend the basic algorithm with further filtering and tracking methods and evaluate their influence on the performance. Finally, we also test our algorithm within a university environment and on a real robot within a home environment.