Dans cette thèse nous intéressons à la conception d'un robot mobile capable d’analyser le comportement et le mouvement d’une personne en environnement intérieur et encombré, par exemple le domicile d’une personne âgée. Plus précisément, notre objectif est de doter le robot des capacités de perception visuelle de la posture humaine de façon à mieux maîtriser certaines situations qui nécessitent de comprendre l’intention des personnes avec lesquelles le robot interagit, ou encore de détecter des situations à risques comme les chutes ou encore d’analyser les capacités motrices des personnes dont il a la garde. Le suivi de la posture dans un environnement dynamique et encombré relève plusieurs défis notamment l'apprentissage en continue du fond de la scène et l'extraction la silhouette qui peut être partiellement observable lorsque la personne est dans des endroits occultés. Ces difficultés rendent le suivi de la posture une tâche difficile. La majorité des méthodes existantes, supposent que la scène est statique et la personne est toujours visible en entier. Ces approches ne sont pas adaptées pour fonctionner dans des conditions réelles. Nous proposons, dans cette thèse, un nouveau système de suivi capable de suivre la posture de la personne dans ces conditions réelles. Notre approche utilise une grille d'occupation avec un modèle de Markov caché pour apprendre en continu l'évolution de la scène et d'extraire la silhouette, ensuite un algorithme de filtrage particulaire hiérarchique est utilisé pour reconstruire la posture. Nous proposons aussi un nouvel algorithme de gestion d'occlusion capable d'identifier et d'exclure les parties du corps cachées du processus de l'estimation de la pose. Finalement, nous avons proposé une base de données contenant des images RGB-D avec la vérité-terrain dans le but d'établir une nouvelle référence pour l'évaluation des systèmes de capture de mouvement dans un environnement réel avec occlusions. La vérité-terrain est obtenue à partir d'un système de capture de mouvement à base de marqueur de haute précision avec huit caméras infrarouges. L'ensemble des données est disponible en ligne. La deuxième contribution de cette thèse, est le développement d'une méthode de localisation visuelle à partir d'une caméra du type RGB-D montée sur un robot qui se déplace dans un environnement dynamique. En effet, le système de capture de mouvement que nous avons développé doit équiper un robot se déplaçant dans une scène. Ainsi, l'estimation de mouvement du robot est importante pour garantir une extraction de silhouette correcte pour le suivi. La difficulté majeure de la localisation d'une caméra dans un environnement dynamique, est que les objets mobiles de la scène induisent un mouvement supplémentaire qui génère des pixels aberrants. Ces pixels doivent être exclus du processus de l'estimation du mouvement de la caméra. Nous proposons ainsi une extension de la méthode de localisation dense basée sur le flux optique pour isoler les pixels aberrants en utilisant l'algorithme de RANSAC. / In this thesis we are interested in designing a mobile robot able to analyze the behavior and movement of a a person in indoor and cluttered environment. Our goal is to equip the robot by visual perception capabilities of the human posture to better analyze situations that require understanding of person with which the robot interacts, or detect risk situations such as falls or analyze motor skills of the person. Motion capture in a dynamic and crowded environment raises multiple challenges such as learning the background of the environment and extracting the silhouette that can be partially observable when the person is in hidden places. These difficulties make motion capture difficult. Most of existing methods assume that the scene is static and the person is always fully visible by the camera. These approaches are not able to work in such realistic conditions. In this thesis, We propose a new motion capture system capable of tracking a person in realistic world conditions. Our approach uses a 3D occupancy grid with a hidden Markov model to continuously learn the changing background of the scene and to extract silhouette of the person, then a hierarchical particle filtering algorithm is used to reconstruct the posture. We propose a novel occlusion management algorithm able to identify and discards hidden body parts of the person from process of the pose estimation. We also proposed a new database containing RGBD images with ground truth data in order to establish a new benchmark for the assessment of motion capture systems in a real environment with occlusions. The ground truth is obtained from a motion capture system based on high-precision marker with eight infrared cameras. All data is available online. The second contribution of this thesis is the development of a new visual odometry method to localize an RGB-D camera mounted on a robot moving in a dynamic environment. The major difficulty of the localization in a dynamic environment, is that mobile objects in the scene induce additional movement that generates outliers pixels. These pixels should be excluded from the camera motion estimation process in order to produce accurate and precise localization. We thus propose an extension of the dense localization method based on the optical flow method to remove outliers pixels using the RANSAC algorithm.

The estimation of motion of multi-camera systems is one of the most important tasks in computer vision research. Recently, some issues have been raised about general camera models and multi-camera systems. Using many cameras as a single camera is studied [60], and the epipolar geometry constraints of general camera models is theoretically derived. Methods for calibration, including a self-calibration method for general camera models, are studied [78, 62]. Multi-camera systems are an example of practically implementable general camera models and they are widely used in many applications nowadays because of both the low cost of digital charge-coupled device (CCD) cameras and the high resolution of multiple images from the wide field of views. To our knowledge, no research has been conducted on the relative motion of multi-camera systems with non-overlapping views to obtain a geometrically optimal solution. ¶ In this thesis, we solve the camera motion problem for multi-camera systems by using linear methods and convex optimization techniques, and we make five substantial and original contributions to the field of computer vision. First, we focus on the problem of translational motion of omnidirectional cameras, which are multi-camera systems, and present a constrained minimization method to obtain robust estimation results. Given known rotation, we show that bilinear and trilinear relations can be used to build a system of linear equations, and singular value decomposition (SVD) is used to solve the equations. Second, we present a linear method that estimates the relative motion of generalized cameras, in particular, in the case of non-overlapping views. We also present four types of generalized cameras, which can be solvable using our proposed, modified SVD method. This is the first study finding linear relations for certain types of generalized cameras and performing experiments using our proposed linear method. Third, we present a linear 6-point method (5 points from the same camera and 1 point from another camera) that estimates the relative motion of multi-camera systems, where cameras have no overlapping views. In addition, we discuss the theoretical and geometric analyses of multi-camera systems as well as certain critical configurations where the scale of translation cannot be determined. Fourth, we develop a global solution under an L∞ norm error for the relative motion problem of multi-camera systems using second-order cone programming. Finally, we present a fast searching method to obtain a global solution under an L∞ norm error for the relative motion problem of multi-camera systems, with non-overlapping views, using a branch-and-bound algorithm and linear programming (LP). By testing the feasibility of LP at the earlier stage, we reduced the time of computation of solving LP.¶ We tested our proposed methods by performing experiments with synthetic and real data. The Ladybug2 camera, for example, was used in the experiment on estimation of the translation of omnidirectional cameras and in the estimation of the relative motion of non-overlapping multi-camera systems. These experiments showed that a global solution using L∞ to estimate the relative motion of multi-camera systems could be achieved.

Thesis (MScEng)--Stellenbosch University, 2012. / ENGLISH ABSTRACT: Simultaneous localization and mapping (SLAM) is vital for autonomous robot navigation. The robot must build a map of its environment while tracking its own motion through that map. Although many solutions to this intricate problem have been proposed, one of the most prominent issues that still needs to be resolved is to accurately measure and track landmarks over time. In this thesis we investigate the use of stereo vision for this purpose. In order to find landmarks in images we explore the use of two feature detectors: the scale-invariant feature transform (SIFT) and speeded-up robust features (SURF). Both these algorithms find salient points in images and calculate a descriptor for each point that is invariant to scale, rotation and illumination. By using the descriptors we match these image features between stereo images and use the geometry of the system to calculate a set of 3D landmark measurements. A Taylor approximation of this transformation is used to derive a Gaussian noise model for the measurements. The measured landmarks are matched to landmarks in a map to find correspondences. We find that this process often incorrectly matches ambiguous landmarks. To find these mismatches we develop a novel outlier detection scheme based on the random sample consensus (RANSAC) framework. We use a similarity transformation for the RANSAC model and derive a probabilistic consensus measure that takes the uncertainties of landmark locations into account. Through simulation and practical tests we find that this method is a significant improvement on the standard approach of using the fundamental matrix. With accurately identified landmarks we are able to perform SLAM. We investigate the use of three popular SLAM algorithms: EKF SLAM, FastSLAM and FastSLAM 2. EKF SLAM uses a Gaussian distribution to describe the systems states and linearizes the motion and measurement equations with Taylor approximations. The two FastSLAM algorithms are based on the Rao-Blackwellized particle filter that uses particles to describe the robot states, and EKFs to estimate the landmark states. FastSLAM 2 uses a refinement process to decrease the size of the proposal distribution and in doing so decreases the number of particles needed for accurate SLAM. We test the three SLAM algorithms extensively in a simulation environment and find that all three are capable of very accurate results under the right circumstances. EKF SLAM displays extreme sensitivity to landmark mismatches. FastSLAM, on the other hand, is considerably more robust against landmark mismatches but is unable to describe the six-dimensional state vector required for 3D SLAM. FastSLAM 2 offers a good compromise between efficiency and accuracy, and performs well overall. In order to evaluate the complete system we test it with real world data. We find that our outlier detection algorithm is very effective and greatly increases the accuracy of the SLAM systems. We compare results obtained by all three SLAM systems, with both feature detection algorithms, against DGPS ground truth data and achieve accuracies comparable to other state-of-the-art systems. From our results we conclude that stereo vision is viable as a sensor for SLAM. / AFRIKAANSE OPSOMMING: Gelyktydige lokalisering en kartering (simultaneous localization and mapping, SLAM) is ’n noodsaaklike proses in outomatiese robot-navigasie. Die robot moet ’n kaart bou van sy omgewing en tegelykertyd sy eie beweging deur die kaart bepaal. Alhoewel daar baie oplossings vir hierdie ingewikkelde probleem bestaan, moet een belangrike saak nog opgelos word, naamlik om landmerke met verloop van tyd akkuraat op te spoor en te meet. In hierdie tesis ondersoek ons die moontlikheid om stereo-visie vir hierdie doel te gebruik. Ons ondersoek die gebruik van twee beeldkenmerk-onttrekkers: scale-invariant feature transform (SIFT) en speeded-up robust features (SURF). Altwee algoritmes vind toepaslike punte in beelde en bereken ’n beskrywer vir elke punt wat onveranderlik is ten opsigte van skaal, rotasie en beligting. Deur die beskrywer te gebruik, kan ons ooreenstemmende beeldkenmerke soek en die geometrie van die stelsel gebruik om ’n stel driedimensionele landmerkmetings te bereken. Ons gebruik ’n Taylor- benadering van hierdie transformasie om ’n Gaussiese ruis-model vir die metings te herlei. Die gemete landmerke se beskrywers word dan vergelyk met dié van landmerke in ’n kaart om ooreenkomste te vind. Hierdie proses maak egter dikwels foute. Om die foutiewe ooreenkomste op te spoor het ons ’n nuwe uitskieterherkenningsalgoritme ontwikkel wat gebaseer is op die RANSAC-raamwerk. Ons gebruik ’n gelykvormigheidstransformasie vir die RANSAC-model en lei ’n konsensusmate af wat die onsekerhede van die ligging van landmerke in ag neem. Met simulasie en praktiese toetse stel ons vas dat die metode ’n beduidende verbetering op die standaardprosedure, waar die fundamentele matriks gebruik word, is. Met ons akkuraat geïdentifiseerde landmerke kan ons dan SLAM uitvoer. Ons ondersoek die gebruik van drie SLAM-algoritmes: EKF SLAM, FastSLAM en FastSLAM 2. EKF SLAM gebruik ’n Gaussiese verspreiding om die stelseltoestande te beskryf en Taylor-benaderings om die bewegings- en meetvergelykings te lineariseer. Die twee FastSLAM-algoritmes is gebaseer op die Rao-Blackwell partikelfilter wat partikels gebruik om robottoestande te beskryf en EKF’s om die landmerktoestande af te skat. FastSLAM 2 gebruik ’n verfyningsproses om die grootte van die voorstelverspreiding te verminder en dus die aantal partikels wat vir akkurate SLAM benodig word, te verminder. Ons toets die drie SLAM-algoritmes deeglik in ’n simulasie-omgewing en vind dat al drie onder die regte omstandighede akkurate resultate kan behaal. EKF SLAM is egter baie sensitief vir foutiewe landmerkooreenkomste. FastSLAM is meer bestand daarteen, maar kan nie die sesdimensionele verspreiding wat vir 3D SLAM vereis word, beskryf nie. FastSLAM 2 bied ’n goeie kompromie tussen effektiwiteit en akkuraatheid, en presteer oor die algemeen goed. Ons toets die hele stelsel met werklike data om dit te evalueer, en vind dat ons uitskieterherkenningsalgoritme baie effektief is en die akkuraatheid van die SLAM-stelsels beduidend verbeter. Ons vergelyk resultate van die drie SLAM-stelsels met onafhanklike DGPS-data, wat as korrek beskou kan word, en behaal akkuraatheid wat vergelykbaar is met ander toonaangewende stelsels. Ons resultate lei tot die gevolgtrekking dat stereo-visie ’n lewensvatbare sensor vir SLAM is.

The construction of fish passes has been a longstanding measure to improve river ecosystem status by ensuring the passability of weirs, dams and other in- channel structures for migratory fish. Many fish passes have a low biological effectiveness because of unsuitable hydrodynamic conditions hindering fish to rapidly detect the pass entrance. There has been a need for techniques to quantify the hydrodynamics surrounding fish pass entrances in order to identify those passes that require enhancement and to improve the design of new passes. This PhD thesis presents the development of a methodology for the rapid, spatially continuous quantification of near-pass hydrodynamics in the field. The methodology involves moving-vessel Acoustic Doppler Current Profiler (ADCP) measurements in order to quantify the 3-dimensional water velocity distribution around fish pass entrances. The approach presented in this thesis is novel because it integrates a set of techniques to make ADCP data robust against errors associated with the environmental conditions near engineered in-channel structures. These techniques provide solutions to (i) ADCP compass errors from magnetic interference, (ii) bias in water velocity data caused by spatial flow heterogeneity, (iii) the accurate ADCP positioning in locales with constrained line of sight to navigation satellites, and (iv) the accurate and cost-effective sensor deployment following pre-defined sampling strategies. The effectiveness and transferability of the methodology were evaluated at three fish pass sites covering conditions of low, medium and high discharge. The methodology outputs enabled a detailed quantitative characterisation of the fish pass attraction flow and its interaction with other hydrodynamic features. The outputs are suitable to formulate novel indicators of hydrodynamic fish pass attractiveness and they revealed the need to refine traditional fish pass design guidelines.

No description available.

Submitted by Luciana Ferreira (lucgeral@gmail.com) on 2018-11-21T11:05:44Z No. of bitstreams: 2 Dissertação - Vinícius Araújo Santos - 2018.pdf: 14601054 bytes, checksum: e02a8bcd3cdc93bf2bf202c3933b3f27 (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) / Approved for entry into archive by Luciana Ferreira (lucgeral@gmail.com) on 2018-11-21T11:06:26Z (GMT) No. of bitstreams: 2 Dissertação - Vinícius Araújo Santos - 2018.pdf: 14601054 bytes, checksum: e02a8bcd3cdc93bf2bf202c3933b3f27 (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) / Made available in DSpace on 2018-11-21T11:06:26Z (GMT). No. of bitstreams: 2 Dissertação - Vinícius Araújo Santos - 2018.pdf: 14601054 bytes, checksum: e02a8bcd3cdc93bf2bf202c3933b3f27 (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) Previous issue date: 2018-10-11 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / Visual Odometry is an important process in image based navigation of robots. The standard methods of this field rely on the good feature matching between frames where feature detection on images stands as a well adressed problem within Computer Vision. Such techniques are subject to illumination problems, noise and poor feature localization accuracy. Thus, 3D information on a scene may mitigate the uncertainty of the features on images. Deep Learning techniques show great results when dealing with common difficulties of VO such as low illumination conditions and bad feature selection. While Visual Odometry and Deep Learning have been connected previously, no techniques applying Siamese Convolutional Networks on depth infomation given by disparity maps have been acknowledged as far as this work’s researches went. This work aims to fill this gap by applying Deep Learning to estimate egomotion through disparity maps on an Siamese architeture. The SiameseVO-Depth architeture is compared to state of the art techniques on OV by using the KITTI Vision Benchmark Suite. The results reveal that the chosen methodology succeeded on the estimation of Visual Odometry although it doesn’t outperform the state-of-the-art techniques. This work presents fewer steps in relation to standard VO techniques for it consists of an end-to-end solution and demonstrates a new approach of Deep Learning applied to Visual Odometry. / Odometria Visual é um importante processo na navegação de robôs baseada em imagens. Os métodos clássicos deste tema dependem de boas correspondências de características feitas entre imagens sendo que a detecção de características em imagens é um tema amplamente discutido no campo de Visão Computacional. Estas técnicas estão sujeitas a problemas de iluminação, presença de ruído e baixa de acurácia de localização. Nesse contexto, a informação tridimensional de uma cena pode ser uma forma de mitigar as incertezas sobre as características em imagens. Técnicas de Deep Learning têm demonstrado bons resultados lidando com problemas comuns em técnicas de OV como insuficiente iluminação e erros na seleção de características. Ainda que já existam trabalhos que relacionam Odometria Visual e Deep Learning, não foram encontradas técnicas que utilizem Redes Convolucionais Siamesas com sucesso utilizando informações de profundidade de mapas de disparidade durante esta pesquisa. Este trabalho visa preencher esta lacuna aplicando Deep Learning na estimativa do movimento por de mapas de disparidade em uma arquitetura Siamesa. A arquitetura SiameseVO-Depth proposta neste trabalho é comparada à técnicas do estado da arte em OV utilizando a base de dados KITTI Vision Benchmark Suite. Os resultados demonstram que através da metodologia proposta é possível a estimativa dos valores de uma Odometria Visual ainda que o desempenho não supere técnicas consideradas estado da arte. O trabalho proposto possui menos etapas em comparação com técnicas clássicas de OV por apresentar-se como uma solução fim-a-fim e apresenta nova abordagem no campo de Deep Learning aplicado à Odometria Visual.

L’objectif de la thèse est de développer un nouveau système de localisation mobile composé de trois caméras 3D de type Kinect et d’une caméra additionnelle de type Fish Eye. La solution algorithmique est basée sur l’odométrie visuelle et permet de calculer la trajectoire du mobile en temps réel à partir des données fournies par les caméras 3D. L’originalité de la méthodologie réside dans l’exploitation d’orthoimages créées à partir des nuages de points acquis en temps réel par les trois caméras. L’étude des différences entre les orthoimages successives acquises par le système mobile permet d’en déduire ses positions successives et d’en calculer sa trajectoire. / The aim of the thesis is to develop a new kind of localization system, composed of three 3D cameras such as Kinect and an additional Fisheye camera. The localization algorithm is based on Visual Odometry principles in order to calculate the trajectory of the mobile platform in real time from the data provided by the 3D cameras.The originality of the processing method lies within the exploitation of orthoimages processed from the point clouds that are acquired in real time by the three cameras. The positions and trajectory of the mobile platform can be derived from the study of the differences between successive orthoimages.

This thesis describes the design and implementation of system for vehicle speed estimation from on-board camera recording. Speed estimation is based on optical flow estimation and convolutional neural network. Designed system is able to estimate speed with average error of 20% on created data set where actual speed is greater than 35 kilometers per hour.

Les systèmes d'aide à la conduite et les travaux concernant le véhicule autonome ont atteint une certaine maturité durant ces dernières aimées grâce à l'utilisation de technologies avancées. Une étape fondamentale pour ces systèmes porte sur l'estimation du mouvement et de la structure de l'environnement (Structure From Motion) pour accomplir plusieurs tâches, notamment la détection d'obstacles et de marquage routier, la localisation et la cartographie. Pour estimer leurs mouvements, de tels systèmes utilisent des capteurs relativement chers. Pour être commercialisés à grande échelle, il est alors nécessaire de développer des applications avec des dispositifs bas coûts. Dans cette optique, les systèmes de vision se révèlent une bonne alternative. Une nouvelle méthode basée sur des approches 2D/2D à partir d'un réseau de caméras asynchrones est présentée afin d'obtenir le déplacement et la structure 3D à l'échelle absolue en prenant soin d'estimer les facteurs d'échelle. La méthode proposée, appelée méthode des triangles, se base sur l'utilisation de trois images formant un triangle : deux images provenant de la même caméra et une image provenant d'une caméra voisine. L'algorithme admet trois hypothèses: les caméras partagent des champs de vue communs (deux à deux), la trajectoire entre deux images consécutives provenant d'une même caméra est approximée par un segment linéaire et les caméras sont calibrées. La connaissance de la calibration extrinsèque entre deux caméras combinée avec l'hypothèse de mouvement rectiligne du système, permet d'estimer les facteurs d'échelle absolue. La méthode proposée est précise et robuste pour les trajectoires rectilignes et présente des résultats satisfaisants pour les virages. Pour affiner l'estimation initiale, certaines erreurs dues aux imprécisions dans l'estimation des facteurs d'échelle sont améliorées par une méthode d'optimisation : un ajustement de faisceaux local appliqué uniquement sur les facteurs d'échelle absolue et sur les points 3D. L'approche présentée est validée sur des séquences de scènes routières réelles et évaluée par rapport à la vérité terrain obtenue par un GPS différentiel. Une application fondamentale dans les domaines d'aide à la conduite et de la conduite automatisée est la détection de la route et d'obstacles. Pour un système asynchrone, une première approche pour traiter cette application est présentée en se basant sur des cartes de disparité éparses. / Driver assistance systems and autonomous vehicles have reached a certain maturity in recent years through the use of advanced technologies. A fundamental step for these systems is the motion and the structure estimation (Structure From Motion) that accomplish several tasks, including the detection of obstacles and road marking, localisation and mapping. To estimate their movements, such systems use relatively expensive sensors. In order to market such systems on a large scale, it is necessary to develop applications with low cost devices. In this context, vision systems is a good alternative. A new method based on 2D/2D approaches from an asynchronous multi-camera network is presented to obtain the motion and the 3D structure at the absolute scale, focusing on estimating the scale factors. The proposed method, called Triangle Method, is based on the use of three images forming a. triangle shape: two images from the same camera and an image from a neighboring camera. The algorithrn has three assumptions: the cameras share common fields of view (two by two), the path between two consecutive images from a single camera is approximated by a line segment, and the cameras are calibrated. The extrinsic calibration between two cameras combined with the assumption of rectilinear motion of the system allows to estimate the absolute scale factors. The proposed method is accurate and robust for straight trajectories and present satisfactory results for curve trajectories. To refine the initial estimation, some en-ors due to the inaccuracies of the scale estimation are improved by an optimization method: a local bundle adjustment applied only on the absolute scale factors and the 3D points. The presented approach is validated on sequences of real road scenes, and evaluated with respect to the ground truth obtained through a differential GPS. Finally, another fundamental application in the fields of driver assistance and automated driving is road and obstacles detection. A method is presented for an asynchronous system based on sparse disparity maps

This Master thesis tackles the problem of measuring objects’ directions from a motionless observation point. A new method based on a single rotating camera requiring the knowledge of only two (or more) landmarks’ direction is proposed. In a first phase, multi-view geometry is used to estimate camera rotations and key elements’ direction from a set of overlapping images. Then in a second phase, the direction of any object can be estimated by resectioning the camera associated to a picture showing this object. A detailed description of the algorithmic chain is given, along with test results on both synthetic data and real images taken with an infrared camera. / Detta masterarbete behandlar problemet med att mäta objekts riktningar från en fast observationspunkt. En ny metod föreslås, baserad på en enda roterande kamera som kräver endast två (eller flera) landmärkens riktningar. I en första fas används multiperspektivgeometri, för att uppskatta kamerarotationer och nyckelelements riktningar utifrån en uppsättning överlappande bilder. I en andra fas kan sedan riktningen hos vilket objekt som helst uppskattas genom att kameran, associerad till en bild visande detta objekt, omsektioneras. En detaljerad beskrivning av den algoritmiska kedjan ges, tillsammans med testresultat av både syntetisk data och verkliga bilder tagen med en infraröd kamera.