One of the main goals in computer vision is to achieve a human-like understanding of images. Nevertheless, image understanding has been mainly studied in the 2D image frame, so more information is needed to relate them to the 3D world. With the emergence of 3D sensors (e.g. the Microsoft Kinect), which provide depth along with color information, the task of propagating 2D knowledge into 3D becomes more attainable and enables interaction between a machine (e.g. robot) and its environment. This dissertation focuses on three aspects of indoor 3D scene understanding: (1) 2D-driven 3D object detection for single frame scenes with inherent 2D information, (2) 3D object instance segmentation for 3D reconstructed scenes, and (3) using room and floor orientation for automatic labeling of indoor scenes that could be used for self-supervised object segmentation. These methods allow capturing of physical extents of 3D objects, such as their sizes and actual locations within a scene.

Cette thèse traite du problème de la Localisation et Cartographie Simultanées (SLAM) dans les environnements d’intérieur. Dans ce contexte, nous avons choisi un SLAM visuel en utilisant les données d’un capteur RGB-D de type Kinect pour estimer la trajectoire de la caméra et construire une carte 3D de l’environnement en temps réel. Malgré les avantages des caméras RGB-D (faible coût, images couleurs et cartes de profondeur), les données de profondeur issues de ce genre de capteur peuvent être de mauvaise qualité ce qui affecte l’estimation de la pose. En outre, la taille des nuages de points engendre une carte globale lourde et contenant de nombreux points 3D redondants. Afin de diminuer l’impact de ces faiblesses sur la résolution du SLAM, nous proposons d’utiliser des plans 3D, majoritaires dans les scènes d’intérieur, dans le processus d’estimation de poses de la caméra pour construire des cartes 3D basées-plans.Les plans 3D servent alors à générer des points d’intérêt 3D planaires moins bruités que les points bruts déduits directement des nuages de points. En rectifiant les valeurs de profondeur des points d’intérêt 3D bruts appartenant à ces plans, nous améliorons ainsi l’estimation de pose quand la scène est composée essentiellement de plans. Par la suite, les plans 3D détectés sont utilisés pour construire une carte 3D globale légère. La carte est élaborée en fusionnant itérativement les régions planaires détectées dans la scène avec celles déjà présentes dans la carte ou en ajoutant de nouveaux plans. Contrairement à la représentation classique basée point,nous réduisons ainsi la taille de la carte 3D et construisons des cartes compactes. Ces cartes sont exploitables par des applications de robotique mobile et de navigation. Pour montrer les bénéfices des travaux proposés dans cette thèse, les expérimentations réalisées évaluent la précision de la localisation, l’influence de l’échantillonnage des données RGB-D sur la détection des plans ainsi que la qualité de la carte basée-plans 3D par rapport à la scène réelle. La carte ainsi constituée de plans présente une première étape vers une carte plus sémantique. / This thesis deals with the Simultaneous Localisation and Mapping (SLAM) problem in indoor environments. In this context, we chose a visual SLAM using an RGB-D sensor (Kinect) to estimate the camera trajectory and to build a 3D map of the environment in realtime. Despite RGB-D cameras advantages (low cost, color images and depth maps), depth data resulting from this kind of sensors may be noisy, which affects pose estimation. In addition,due to points clouds sizes, the resulting global map is heavyweight and contains many redundant 3D points. In order to reduce the impact of these weaknesses on resolving the SLAM problem, we propose to use 3D planes, which are dominant in indoor scenes, for both camera poses estimations and 3D based-planes maps building process. Hence, 3D planes are used to generate 3D planar feature featuring less depth noise than the raw points extracted directly from points clouds. By regularizing depth values of raw 3D feature points belonging to these planes, we improve pose estimation when the scene is mainly composed of planes. Then, the detected 3D planes are used to build the global 3D map, creating a light representation of the environment based on these planes. The map is iteratively built from each new camera pose either by merging new planes to the existing ones or by adding new planes to the map. Thus,unlike conventional point-based representation, the size of the resulting 3D map is considerably reduced and the built map is more compact compared to point-based maps.These maps maybe used by mobile robotics and navigation applications. To show the benefits of our works, the conducted experiments to evaluate localisation accuracy, the influence of subsampled RGB-Ddata on plane detection, as well as quality of 3D plane-based maps against real scenes. Such plane-based maps represents a first step towards semantic maps.

This diploma thesis is devoted to the use of 3D reconstruction using EBSD method for microstructural analysis of silicon nitride ceramic material predetermined for the rolling-mill used in the wire production. Application of this method can be used for the grain structure reconstruction and basic microstructural parameters can be than extracted. The development of a suitable method for 3D reconstruction of the structure of the materials the main aim of this work. There are described the different steps begun by sample preparation from the investigated material trough the optimisation of analysis parameters up to the visualization of the grain structure. New type of sample geometry has been designed which leads to the better and faster observation of the microstructure of ceramic materials. This thesis also describes optimal reconstruction parameters such as the geometry of the assembly used in the microscope without mechanical movement of the sample or the influence of conductive coating prepared via in-situ sputtering of suitable metal, or adjustment of the electron and ion beams. Further are described two options of software which can be used for the final generation of 3D structure information and are assessed their advantages and disadvantages. The effect of the filter setting and other parameters and their influence on the resulting structural parameters are also evaluated.

Recent research makes use of biologically inspired computation and artificial intelligence as efficient means to solve real-world problems. Humans show a significant performance in extracting and interpreting visual information. In the cases where visual data is not available, or, for example, if it fails to provide comprehensive information due to occlusions, tactile exploration assists in the interpretation and better understanding of the environment. This cooperation between human senses can serve as an inspiration to embed a higher level of intelligence in computational models. In the context of this research, in the first step, computational models of visual attention are explored to determine salient regions on the surface of objects. Two different approaches are proposed. The first approach takes advantage of a series of contributing features in guiding human visual attention, namely color, contrast, curvature, edge, entropy, intensity, orientation, and symmetry are efficiently integrated to identify salient features on the surface of 3D objects. This model of visual attention also learns to adaptively weight each feature based on ground-truth data to ensure a better compatibility with human visual exploration capabilities. The second approach uses a deep Convolutional Neural Network (CNN) for feature extraction from images collected from 3D objects and formulates saliency as a fusion map of regions where the CNN looks at, while classifying the object based on their geometrical and semantic characteristics. The main difference between the outcomes of the two algorithms is that the first approach results in saliencies spread over the surface of the objects while the second approach highlights one or two regions with concentrated saliency. Therefore, the first approach is an appropriate simulation of visual exploration of objects, while the second approach successfully simulates the eye fixation locations on objects. In the second step, the first computational model of visual attention is used to determine scattered salient points on the surface of objects based on which simplified versions of 3D object models preserving the important visual characteristics of objects are constructed. Subsequently, the thesis focuses on the topic of tactile object recognition, leveraging the proposed model of visual attention. Beyond the sensor technologies which are instrumental in ensuring data quality, biological models can also assist in guiding the placement of sensors and support various selective data sampling strategies that allow exploring an object’s surface faster. Therefore, the possibility to guide the acquisition of tactile data based on the identified visually salient features is tested and validated in this research. Different object exploration and data processing approaches were used to identify the most promising solution. Our experiments confirm the effectiveness of computational models of visual attention as a guide for data selection for both simplifying 3D representation of objects as well as enhancing tactile object recognition. In particular, the current research demonstrates that: (1) the simplified representation of objects by preserving visually salient characteristics shows a better compatibility with human visual capabilities compared to uniformly simplified models, and (2) tactile data acquired based on salient visual features are more informative about the objects’ characteristics and can be employed in tactile object manipulation and recognition scenarios. In the last section, the thesis addresses the issue of transfer of learning from vision to touch. Inspired from biological studies that attest similarities between the processing of visual and tactile stimuli in human brain, the thesis studies the possibility of transfer of learning from vision to touch using deep learning architectures and proposes a hybrid CNN that handles both visual and tactile object recognition.

No description available.

Image-based 3D reconstruction has been employed in industrial metrology for micro measurements and quality control purposes. However, generating a highly-detailed and reliable 3D reconstruction of non-collaborative surfaces (textureless, shiny, and transparent) is still an open issue. This thesis presents various methodologies to successfully generate a highly-detailed and reliable 3D reconstruction of non-collaborative objects using the proposed photometric stereo image acquisition system. The first proposed method employs geometric construction to integrate photogrammetry and photometric stereo in order to overcome each technique's limitations and to leverage each technique's strengths in order to reconstruct an accurate and high-resolution topography of non-collaborative surfaces. This method uses accurate photogrammetric 3D measurements to rectify the global shape deviation of photometric stereo meanwhile uses photometric stereo to recover the high detailed topography of the object. The second method combines the high spatial frequencies of photometric stereo depth map with the low frequencies of photogrammetric depth map in frequency domain to produce accurate low frequencies while retaining high frequencies. For the third approach, we utilize light directionality to improve texture quality by leveraging shade and shadow phenomena using the proposed image-capturing system that employs several light sources for highlighting roughness and microstructures on the surface. And finally, we present two methods that effectively orient images by leveraging the low-contrast textures highlighted on object surfaces (roughness and 3D microstructures) using proper lighting system. Various objects with different surface characteristics including textureless, reflective, and transparent are used to evaluate different proposed approaches. To assess the accuracy of each approach, a comprehensive comparison between reference data and generated 3D points is provided.

Magistrantūros baigiamojo darbo tikslas yra išanalizuoti 3D lazerines matavimo sistemas, jų panaudojimą geodeziniams darbams ir palyginti su elektroniniu tacheometru. Atliekant tyrimą buvo panaudoti teoriniai bei praktiniai duomenys ir sisteminė analizė. Tyrimo objektas- stacionarus lazerinis skeneris Riegl LMS-Z420i. Pasirinktas prietaisas yra kol kas vienintelis toks Lietuvoje. Modernios technologijos su naujausiais techniniais sprendimais dėl savo produktyvumo gauna vis didesnę reikšmę geodezijoje. Todėl būtina žinoti jų veikimo principus ir panaudojimo galimybes. 3D lazerinės matavimų sistemos palyginti su elektroniniais tachometrais yra naujas dalykas. 3D lazerinės matavimų sistemos - tai įrenginiai, 3D skeneriai, kurie analizuoja realaus pasaulio objektą ir jo aplinką, siekiant surinkti duomenis apie jo formą ir jo išvaizdą (t.y. spalvą). Surinkti duomenys gali būti naudojami kurti skaitmeninius trimačius modeliu- tai informacija naudinga įvairioms programoms. Šie prietaisai yra plačiai naudojami topografiniuose darbuose, kalnakasyboje, fasadų matavimams, pastatų inventorizacijai, miestų modeliavimui. Taip pat ši technologija apima pramoninio dizaino, protezavimo, prototipų kūrimo, kokybės kontrolės tikrinimo funkcijas. Darbe išanalizuoti lazerinis skeneris Riegl LMS-Z420i, jo parametrai bei funkcijos, matavimo technologija ir palyginta matavimo sparta, tikslumas ir duomenų apdorojimas su elektroniniu tacheometru Sokkia set 5x. Atlikus palyginimą lazerinis skeneris... [toliau žr. visą tekstą] / Magister thesis is to analyze the 3D laser measurement systems, the use of geodetic work and compared with an electronic tachometer having. The study was theoretical and practical use of data and systematic analysis. The object of the stationary-laser scanner Riegl LMS-Z420i. The selected device is so far the only one in Lithuania. Modern technology with the latest technical solutions for its performance are receiving an increasing larger role in geodesy. Therefore, it is necessary to know the principles of operation and usability. 3D laser measurement system in relation to electronic tachometer is a new thing. 3D laser measurement system - a device, 3D scanners, which examines real-world object and its environment in order to collect data about its shape and its appearance (ie color). The data collected can be used to create digital three-dimensional model, this information is useful for various applications. These devices are widely used for topographical works, mining, facade measurements, building inventory, urban simulation. Also, this technology include industrial design, prosthetics, prototype development, quality control inspection functions. The paper analyzed a laser scanner Riegl LMS-Z420i, its parameters and functions, measurement techniques and compared the measurement speed, accuracy and data processing with an electronic tacheometer having Sokkia Set 5x. After a comparison of laser scanner Riegl LMS-Z420i has superior electronic tachometers Sokkia Set 5x.

Thèse numérisée par la Division de la gestion de documents et des archives de l'Université de Montréal.

Thèse numérisée par la Division de la gestion de documents et des archives de l'Université de Montréal

3D-printing, också känt som additiv tillverkning, är en tillverkningsmetod som har revolutionerat många branscher och har växt stort både inom industrin och för privat användning. Tekniken använder sig utav en lager-på-lager metod för att tillverka olika objekt. Med dagens teknik går det att printa ut föremål av exempelvis metall, plast, betong och ett flertal andra material. Additiv tillverkning av betong ger möjligheten att skapa nya smarta konstruktionslösningar, vilket medför stora materialbesparingar och minskat materialspill. Produktionskostnader och hastighet kan också dra nytta av metoden genom att minska arbetskraft och eliminera kostnader för tillverkning och montering av gjutformar. Den här studien bidrar till en ökad förståelse för vad som krävs för att utveckla ett fungerande betongmaterial för additiv tillverkning. För att additiv tillverkning ska kunna standardiseras, bli mer kommersiellt och få en bredare användning krävs en djupare förståelse av betongens materialegenskaper. Detta eftersom materialet skiljer sig från konventionell betong. Syftet med detta projekt är att utveckla en betongblandning anpassad för additiv tillverkning. En undersökning görs för att hitta (i) en betongblandning med lämpliga mekaniska materialegenskaper och (ii) en betongblandning som är väl anpassad till 3Dskrivare. Det viktigaste för att en betongblandning ska kunna användas för additiv tillverkning är att blandningen kan pumpas genom systemet och extraheras genom munstycket vid tillverkning samt att slutmaterialet visar bra byggbarhet. Pumpbarhet är förutsättningen för att betongen ska kunna användas i en 3D-skrivare. Betongen ska vara tillräckligt smidig för att kunna pumpas ut genom ett munstycke, men även ha en tillräckligt god inre sammanhållning för att inte deformeras efter att den har pumpats ut. Pumpbarhet påverkas till stor del av vilken sorts pumpsystem som används. Resultaten varierar beroende på vilken pump, munstycke och slang som används vid materialtesterna. Det förefaller att en generell blandning anpassad för flera olika pumpsystem är svårt att uppnå. I det här arbetet har sex olika blandningar med olika variationer testats. Detta ledde till 38 blandningar som genomgått olika tester. De blandningarna med bäst resultat efter finjusteringar var blandning 4.1 och 5.1, de visade hög kvalité för pumpbarhet och byggbarhet. Blandning 4.1 innehåller vatten, anläggningscement, starvis 3040, glenium, CERW, krossballast och glasfibrer och blandning 5.1 är likadant fast med flygaska istället för CERW. / 3D printing, also known as additive manufacturing, is a manufacturing method that has revolutionized many industries and has grown widely both in industry and private use. The technique means using a layer-upon-layer method to manufacture different objects. With today's technology, it is possible to print objects of, for example, metal, plastic, concrete and several other materials. Additive manufacturing of concrete structures can be used to create new smart design solutions, which means significant material savings and reduced material waste. Production costs and time reduction may also be achieved using the method due to lower labor requirements and reduced costs for manufacturing and assembling of molds. This study contributes to an increased understanding of what is required to develop a functioning concrete material for additive manufacturing. In order for additive manufacturing to be standardized, become commercial and be broadly used, a deeper understanding of the concrete properties is required. This is because the material used in 3D printing differs from conventional concrete. The purpose of this project is to develop a concrete mixture adapted for additive manufacturing. A survey is made to find (i) a concrete mixture with suitable mechanical material properties, and (ii) a concrete mixture well adapted to 3D printers. The most important thing for a concrete mix to be used for additive production is that the mixture can be pumped through the system and extracted through the nozzle during manufacture and that the final material shows good buildability. Pumpability is a prerequisite for the concrete to be used in a 3D printer. The concrete must be sufficiently flexible to be pumped out through a nozzle, but also have a sufficiently good internal cohesion so as not to deform after it has been pumped out. Pumpability is largely affected by the type of pump system used. The results vary depending on the pump, nozzle and hose used in the material tests. It seems that a general mix adapted to several different pump systems is difficult to achieve. In this work, six different mixtures with different variations have been tested. This led to 38 mixtures that underwent various tests. The mixtures with the best results after fine adjustments were mix 4.1 and 5.1, they showed high quality for pumpability and buildability. Mixture 4.1 contains water, plant cement, starvis 3040, glenium, CERW, crush ballast and glass fibers and mixture 5.1 is similarly fixed with fly ash instead of CERW.