Centre dedocumentation
scientifique

Learning stereo reconstruction with deep neural networks / Stepan Tulyakov (2020)  

Public Titre : Learning stereo reconstruction with deep neural networks Type de document : Thèse/HDR Auteurs : Stepan Tulyakov, Auteur ; François Fleuret, Directeur de thèse ; Anton Ivanov, Directeur de thèse Editeur : Lausanne : Ecole Polytechnique Fédérale de Lausanne EPFL Année de publication : 2020 Importance : 139 p. Format : 21 x 30 cm Note générale : bibliographie Thèse présentée à l'Ecole Polytechnique Fédérale de Lausanne pour l’obtention du grade de Docteur ès Sciences Langues : Anglais (eng) Descripteur : [Vedettes matières IGN] Traitement d'image optique [Termes descripteurs IGN] apprentissage profond [Termes descripteurs IGN] classification semi-dirigée [Termes descripteurs IGN] contrainte géométrique [Termes descripteurs IGN] couple stéréoscopique [Termes descripteurs IGN] entropie [Termes descripteurs IGN] estimateur [Termes descripteurs IGN] étalonnage géométrique [Termes descripteurs IGN] modèle stéréoscopique [Termes descripteurs IGN] profondeur [Termes descripteurs IGN] réalité de terrain [Termes descripteurs IGN] reconstruction 3D [Termes descripteurs IGN] reconstruction d'image [Termes descripteurs IGN] vision par ordinateur [Termes descripteurs IGN] vision stéréoscopique Résumé : (auteur) Stereo reconstruction is a problem of recovering a 3d structure of a scene from a pair of images of the scene, acquired from different viewpoints. It has been investigated for decades and many successful methods were developed. The main drawback of these methods, is that they typically utilize a single depth cue, such as parallax, defocus blur or shading, and thus are not as robust as a human visual system that simultaneously relies on a range of monocular and binocular cues. This is mainly because it is hard to manually design a model, accounting for multiple depth cues. In this work, we address this problem by focusing on deep learning-based stereo methods that can discover a model for multiple depth cues directly from training data with ground truth depth. The complexity of deep learning-based methods, however, requires very large training sets with ground truth depth, which is often hard or costly to collect. Furthermore, even when training data is available it is often contaminated with noise, which reduces the effectiveness of supervised learning. In this work, in Chapter 3 we show that it is possible to alleviate this problem by using weakly supervised learning, that utilizes geometric constraints of the problem instead of ground truth depth. Besides the large training set requirement, deep stereo methods are not as application-friendlyas traditional methods. They have a large memory footprint and their disparity range is fixed at training time. For some applications, such as satellite stereo i magery, these are serious problems since satellite images are very large, often reaching tens of megapixels, and have a variable baseline, depending on a time difference between stereo images acquisition. In this work, in Chapter 4 we address these problems by introducing a novel network architecture with a bottleneck, capable of processing large images and utilizing more context, and an estimator that makes the network less sensitive to stereo matching ambiguities and applicable to any disparity range without re-training. Because deep learning-based methods discover depth cues directly from training data, they can be adapted to new data modalities without large modifications. In this work, in Chapter 5 we show that our method, developed for a conventional frame-based camera, can be used with a novel event-based camera, that has a higher dynamic range, smaller latency, and low power consumption. Instead of sampling intensity of all pixels with a fixed frequency, this camera asynchronously reports events of significant pixel intensity changes. To adopt our method to this new data modality, we propose a novel event sequence embedding module, that firstly aggregates information locally, across time, using a novel fully-connected layer for an irregularly sampled continuous domain, and then across discrete spatial domain. One interesting application of stereo is a reconstruction of a planet’s surface topography from satellite stereo images. In this work, in Chapter 6 we describe a geometric calibration method, as well as mosaicing and stereo reconstruction tools that we developed in the framework of the doctoral project for Color and Stereo Surface Imaging System onboard of ESA’s Trace Gas Orbiter, orbiting Mars. For the calibration, we propose a novel method, relying on starfield images because large focal lengths and complex optical distortion of the instrument forbid using standard methods. Scientific and practical results of this work are widely used by a scientific community. Note de contenu : 1- Introduction 2- Background 3- Weakly supervised learning of deep patch-matching cost 4- Applications-friendly deep stereo 5- Dense deep event-based stereo 6- Calibration of a satellite stereo system 7- Conclusions Numéro de notice : 25795 Affiliation des auteurs : non IGN Thématique : IMAGERIE Nature : Thèse étrangère Note de thèse : Thèse de Doctorat : Sciences : Lausanne : 2020 En ligne : https://infoscience.epfl.ch/record/275342?ln=fr Format de la ressource électronique : URL Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=95025

Machine learning and geographic information systems for large-scale mapping of renewable energy potential / Dan Assouline (2019)  

Public Titre : Machine learning and geographic information systems for large-scale mapping of renewable energy potential Type de document : Thèse/HDR Auteurs : Dan Assouline, Auteur ; Jean-Louis Scartezzini, Directeur de thèse ; Nahid Mohajeri Pour Rayeni, Directeur de thèse Editeur : Lausanne : Ecole Polytechnique Fédérale de Lausanne EPFL Année de publication : 2019 Importance : 294 p. Format : 21 x 30 cm Note générale : bibliographie Thèse présentée pour l'obtention du grade de Docteur ès Sciences à l'Ecole Polytechnique Fédérale de Lausanne Langues : Anglais (eng) Descripteur : [Vedettes matières IGN] Géomatique [Termes descripteurs IGN] apprentissage automatique [Termes descripteurs IGN] carte thématique [Termes descripteurs IGN] classification dirigée [Termes descripteurs IGN] classification par forêts aléatoires [Termes descripteurs IGN] classification par séparateurs à vaste marge [Termes descripteurs IGN] données météorologiques [Termes descripteurs IGN] données topographiques [Termes descripteurs IGN] énergie éolienne [Termes descripteurs IGN] énergie géothermique [Termes descripteurs IGN] énergie renouvelable [Termes descripteurs IGN] énergie solaire [Termes descripteurs IGN] méthode fondée sur le noyau [Termes descripteurs IGN] modélisation spatio-temporelle [Termes descripteurs IGN] occupation du sol [Termes descripteurs IGN] prédiction [Termes descripteurs IGN] Suisse Résumé : (auteur) A promising pathway to follow in order to reach sustainable development goals is an increased reliance on renewable sources of energy. The optimized use of these energy sources, however, requires the assessment of their potential supply, along with the demand loads in locations of interest. In particular, large-scale supply estimation studies are needed in order to evaluate areas of high potential for each type of energy source for a particular region, and allow for the elaboration of efficient global energy strategies. In Switzerland, the “Energy Strategy 2050”, initiated in 2011 by the Swiss Federal Council, sets an example with the ambitious goal of reaching a 50-80% reduction of CO2 emissions by the year 2050, with a clear course of action: phasing-out nuclear power, improving energy efficiency, and greatly increasing the use of renewables. This thesis develops a general data-driven strategy combining Geographic Information Systems and Machine Learning methods to map the large-scale energy potential for three very popular sources of decentralized energy systems: wind energy (using horizontal axis wind turbines), geothermal energy (using very shallow ground source heat pumps) and solar energy (using photovoltaic solar panels over rooftops). For each of the three considered energy sources, an adapted methodology is suggested to assess its large-scale potential, by estimating multiple variables of interest (with a suitable time resolution, e.g. monthly or yearly), using widely available data, and combining these variables into potential values. These latter estimated variables, dictating the potential, include: (i) the monthly wind speed, and rural and urban topographic/obstacle configuration for wind energy, (ii) the ground thermal conductivity, volumetric heat capacity and monthly temperature gradient for geothermal energy, (iii) the monthly solar radiation, available area for PV panels over rooftops, geometrical characteristics of rooftops and monthly shading factors over rooftops for solar energy. The use of Machine Learning algorithms (notably Support Vector Machines and Random Forests) allows, given adequate features and training data (examples for some locations), for the prediction of the latter variables at unknown locations, along with the uncertainty attached to the predictions. In each case, the developed methodology is set-up with an aim to be applied for Switzerland, meaning that it relies on Swiss available energy-related data. Such data, however, including meteorological, topographic, ground/soil-related and building-related data, is becoming progressively available for most countries, making it possible to widely generalize the proposed methodologies. Results show that Machine Learning is adequate for energy potential estimation, as the multiple required predictions and spatial extrapolations are achieved with reasonable accuracy. In addition, final values are validated with other existing data or studies when possible, and show general agreement. The application of the suggested potential methodologies in Switzerland outline the very significant potential for the considered renewables. In particular, there is a relatively high potential for RooftopMounted solar PV panels, as it is estimated that they could generate a total electricity production of 16.3 TWh per year, which corresponds to 25.3% of the annual electricity demand in 2017.In each case, the developed methodology is set-up with an aim to be applied for Switzerland, meaning that it relies on Swiss available energy-related data. Such data, however, including meteorological, topographic, ground/soil-related and building-related data, is becoming progressively available for most countries, making it possible to widely generalize the proposed methodologies. Results show that Machine Learning is adequate for energy potential estimation, as the multiple required predictions and spatial extrapolations are achieved with reasonable accuracy. In addition, final values are validated with other existing data or studies when possible, and show general agreement. The application of the suggested potential methodologies in Switzerland outline the very significant potential for the considered renewables. In particular, there is a relatively high potential for RooftopMounted solar PV panels, as it is estimated that they could generate a total electricity production of 16.3 TWh per year, which corresponds to 25.3% of the annual electricity demand in 2017. Note de contenu : 1- Introduction 2- Machine Learning 3- Theory and modeling of renewable energy systems 4- Wind energy: a theoretical potential estimation 5- Very shallow geothermal energy: a theoretical potential estimation 6- Solar energy: a technical potential estimation at commune scale 7- Solar energy: an improved potential estimation at pixel scale 8- Conclusion Numéro de notice : 25797 Affiliation des auteurs : non IGN Thématique : GEOMATIQUE Nature : Thèse française Note de thèse : Thèse de Doctorat : Sciences : EPFLausanne : 2019 En ligne : https://infoscience.epfl.ch/record/264971?ln=fr Format de la ressource électronique : URL Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=95038

Parameter estimation with GNSS-reflectometry and GNSS synthetic aperture techniques / Miguel Angel Ribot Sanfelix (2018)  

Public Titre : Parameter estimation with GNSS-reflectometry and GNSS synthetic aperture techniques Type de document : Thèse/HDR Auteurs : Miguel Angel Ribot Sanfelix, Auteur ; Pierre-André Farine, Directeur de thèse ; Cyril Botteron, Directeur de thèse Editeur : Lausanne : Ecole Polytechnique Fédérale de Lausanne EPFL Année de publication : 2018 Importance : 187 p. Format : 21 x 30 cm Note générale : bibliographie Thèse présentée à l'Ecole Polytechnique Fédérale de Lausanne pour l'obtention du grade de Docteur ès Sciences Langues : Anglais (eng) Descripteur : [Vedettes matières IGN] Traitement du signal [Termes descripteurs IGN] classification par maximum de vraisemblance [Termes descripteurs IGN] correction du trajet multiple [Termes descripteurs IGN] erreur de positionnement [Termes descripteurs IGN] filtre de Kalman [Termes descripteurs IGN] interférence [Termes descripteurs IGN] ligne de visée [Termes descripteurs IGN] phase [Termes descripteurs IGN] récepteur GNSS [Termes descripteurs IGN] réflectométrie par GNSS [Termes descripteurs IGN] signal GNSS [Termes descripteurs IGN] signal GPS Résumé : (auteur) Aside from intentional interference, multipath is the most significant error source for Global Navigation Satellite Systems (GNSS) receivers in many operational scenarios. In this thesis, we study the multipath estimation from two different perspectives: to retrieve useful information from it using GNSS-Reflectometry (GNSS-R) techniques; and to mitigate its effects or to estimate its direction-of-arrival (DOA) as well as the line-of-sight (LOS) signal’s using synthetic aperture (SA) processing. The first part of the thesis focuses on precision bounds for GNSS-R techniques for groundbased receivers, in scenarios where a single antenna simultaneously receives the LOS signal and a specular reflection. First, we derive the Cramér-Rao bound (CRB) of the receiver’s height and the reflection coefficient, with the latter depending on the surface’s electrical properties. More specifically, we propose a CRB derivation applicable to GNSS-R techniques that make use of the phase information and long observation times, such as the interference pattern technique (IPT). The derivation is based on the parameter transformation of the Fisher information matrix. We study the dependence of the computed CRB on the scenario and the receiver bandwidth. The CRB results for the simulated scenarios are consistent with the precision reported for many GNSS-R techniques used in these scenarios. The proposed CRB is meant to benchmark and compare new and existing techniques. Besides the derived CRB, we propose an algorithm to obtain the maximum-likelihood (ML) estimator of the parameters of interest with the IPT: the segmented ML estimator (SML). The SML transforms a complex multivariate optimization problem into multiple simpler ones by dividing the parameter search space taking advantage of the cost function’s particular structure. The SML is validated with simulated signal and asymptotically cross-validates the CRB results. The second part of the thesis is devoted to the study of the SA processing of GNSS signals. The goal is to estimate the DOA of the signals received, and mitigate errors in the navigation solution caused by interfering signals, such as multipath. We start by deriving the CRB for the SA context, as a function of the antenna trajectory. This CRB considers the effect of the antenna complex gain, and we show in simulations that it is possible to achieve meaningful DOA estimation only by changing the antenna’s orientation. We continue by proposing a development framework built upon a signal tracking architecture integrating SA processing. Before any SA processing, it is necessary to estimate and compensate any carrier phase contribution not related to the antenna motion. To do so, we propose two new sequential techniques based on the extended Kalman filter (EKF). Also, we develop an open-loop version of the proposed SA tracking architecture, more robust than its closed-loop counterpart. Finally, we validate the proposed architecture and SA-based techniques with synthetic GPS signals at first, and then with real signals, recorded using an antenna mounted on a mechanical rotating arm. The obtained results validate the implemented techniques and show how the proposed SA architecture can ultimately mitigate the position bias error observed in environments with severe multipath interference. Note de contenu : 1- Introduction 2- GNSS-Reflectomery and Synthetic Aperture Processing: An Overview 3- Parameter Estimation in GNSS-Reflectometry Scenarios with Coherent Reflection 4- Spatial Filtering of GNSS Signals with Synthetic Aperture Processing 5- Conclusions Numéro de notice : 25793 Affiliation des auteurs : non IGN Thématique : IMAGERIE Nature : Thèse étrangère Note de thèse : Thèse de Doctorat : Sciences : Lausanne : 2018 En ligne : https://infoscience.epfl.ch/record/253111?ln=fr Format de la ressource électronique : URL Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=95019

Toward a systematic integration of optical remote sensing for inland waters studies / Vincent Maurice Nouchi (2018)  

Public Titre : Toward a systematic integration of optical remote sensing for inland waters studies Type de document : Thèse/HDR Auteurs : Vincent Maurice Nouchi, Auteur ; Alfred Johny Wüest, Auteur ; Damien Bouffard, Auteur Editeur : Lausanne : Ecole Polytechnique Fédérale de Lausanne EPFL Année de publication : 2018 Importance : 122 p. Format : 21 x 30 cm Note générale : bibliographie Thèse présentée à l'Ecole Polytechnique Fédérale de Lausanne pour l'obtention du grade de Docteur ès Sciences Langues : Anglais (eng) Descripteur : [Vedettes matières IGN] Applications de télédétection [Termes descripteurs IGN] aide à la décision [Termes descripteurs IGN] calcaire [Termes descripteurs IGN] correction atmosphérique [Termes descripteurs IGN] décomposition empirique du signal [Termes descripteurs IGN] hydrodynamique [Termes descripteurs IGN] image Landsat-8 [Termes descripteurs IGN] image MODIS [Termes descripteurs IGN] image Sentinel-MSI [Termes descripteurs IGN] LEMAN (LAC) [Termes descripteurs IGN] modèle dynamique [Termes descripteurs IGN] précipitation [Termes descripteurs IGN] réflectance [Termes descripteurs IGN] variation saisonnière Résumé : (auteur) Freshwater resources play a central role in social and economic development of modern civilisations, yet their value is often underestimated and neglected in developed countries. In fact, freshwater habitats are facing unprecedented threat because of human activities, and it is necessary to provide reliable water quality indicators to monitor the response of aquatic systems. In this context, remote sensing has a great potential to provide a complementary source of data for monitoring and understanding the processes involved in inland waters around the world at fine temporal and spatial resolutions. The scientific approach adopted in this thesis is based on the integration of complementary sources of information provided by state-of-the-art monitoring methods to foster our understanding of freshwater habitats. Specifically, we demonstrate the additional value provided by combining complementary sensors with bio-geochemical measurements and hydrodynamic models, using a rare event in Lake Geneva which got a wide public attention in local newspapers: a calcite precipitation event. The principal focus of the remote-sensing community has recently been directed towards very turbid waters in order to address the challenges involved with the retrieval of mixed constituent concentrations. In this thesis, I highlight some important challenges relative to clearer waters which also require further attention from the community. Specifically, I provide a solution to account for vertical nonuniformities of water constituent concentrations using simple approximation models in Lake Geneva. Finally, I provide a comprehensive comparison between state-of-the-art atmospheric correction methods, which are presumably relevant for inland water monitoring and applicable to the new constellation of remote sensors. The aim is to provide reliable recommendations to help forthcoming studies to apply the most suited procedure to their investigation. Note de contenu : 1- Introduction 2- Resolving biogeochemical processes in lakes using remote sensing 3- Effects of non-uniform vertical constituent profiles on remote-sensing reflectance of oligo- to mesotrophic lakes 4- Inter-comparison of atmospheric corrections for S-2 observations over Lake Geneva 5- Conclusion Numéro de notice : 25790 Affiliation des auteurs : non IGN Thématique : IMAGERIE Nature : Thèse étrangère Note de thèse : Thèse de Doctorat : Sciences : Lausanne : Suisse : 2018 En ligne : https://infoscience.epfl.ch/record/255664?ln=fr Format de la ressource électronique : URL Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=95007

Vehicle dynamic model based navigation for small UAVs / Mehran Khaghani (2018)  

Public Titre : Vehicle dynamic model based navigation for small UAVs Type de document : Thèse/HDR Auteurs : Mehran Khaghani, Auteur ; Jan Skaloud, Directeur de thèse Editeur : Zurich : Schweizerischen Geodatischen Kommission / Commission Géodésique Suisse Année de publication : 2018 Autre Editeur : Lausanne : Ecole Polytechnique Fédérale de Lausanne EPFL Collection : Geodätisch-Geophysikalische Arbeiten in der Schweiz, ISSN 0257-1722 num. 101 Importance : 138 p. Format : 21 x 30 cm ISBN/ISSN/EAN : 978-3-908440-47-5 Note générale : bibliography Thèse de Doctorat, EPFL, Lausanne, 2018 Langues : Anglais (eng) Descripteur : [Vedettes matières IGN] Navigation et positionnement [Termes descripteurs IGN] analyse de sensibilité [Termes descripteurs IGN] centrale inertielle [Termes descripteurs IGN] démonstration de faisabilité [Termes descripteurs IGN] drone [Termes descripteurs IGN] filtre de Kalman [Termes descripteurs IGN] GPS-INS [Termes descripteurs IGN] méthode de Monte-Carlo [Termes descripteurs IGN] modèle de simulation [Termes descripteurs IGN] modèle dynamique [Termes descripteurs IGN] navigation autonome [Termes descripteurs IGN] positionnement par GNSS [Termes descripteurs IGN] processus stochastique [Termes descripteurs IGN] vent Index. décimale : 30.70 Navigation et positionnement Résumé : (auteur) The dominant navigation system for small civilian UAVs today is based on integration of inertial navigation system (INS) and global navigation satellite system (GNSS). This strategy works well to navigate the UAV, as long as proper reception of GNSS signal is maintained. However, when GNSS outage occurs, the INS-based navigation solution drifts very quickly, considering the limited quality of IMU(s) employed in INS for small UAVs. In beyond visual line of sight (BVLOS) flights, this poses the serious danger of losing the UAV and its eventual falling down. Limited payload capacity and cost for small UAVs, as well as the need for operating in different conditions, with limited visibility for example, make it challenging to find a solution to reach higher levels of navigation autonomy based on conventional approaches. This thesis aims to improve the accuracy of autonomous navigation for small UAVs by at least one order of magnitude. The proposed novel approach employs vehicle dynamic model (VDM) as process model within navigation system, and treats data from other sensors such as IMU, barometric altimeter, and GNSS receiver, whenever available, as observations within the system. Such improvement comes with extra effort required to determine the VDM parameters for any specific UAV. This work investigates the internal capability of the proposed system for estimating VDM parameters as part of the augmented state vector within an extended Kalman filter (EKF) as the estimator. This reduces the efforts required to setup such navigation system that is platform dependent. Multiple experimental flights using two custom made fixed-wing UAVs are presented together with Monte-Carlo simulations. The results reveal improvements of 1 to 2 orders of magnitude in navigation accuracy during GNSS outages of a few minutes' duration. Computational cost for the proposed VDM-based navigation does not exceed 3~times that of conventional INS-based systems, which establishes its applicability for online application. A global sensitivity analysis is presented, spotting the VDM parameters with higher influence on navigation performance. This provides insight for design of calibration procedures. The proposed VDM-based navigation system can be interesting for professional UAVs from at least two points of view. Firstly, it adds little to no extra hardware and cost to the UAV. Secondly and more importantly, it might be currently the only way to reach such significant improvement in navigation autonomy for small UAVs regardless of visibility conditions and electromagnetic signals reception. Possibly, such environmental condition independence for navigation system may be needed to obtain certifications from legal authorities to expand UAV applications to new types of mission. Note de contenu : 1- Preliminaries 2- VDM-based navigation framework 3- Results and analyses 4- Conclusion remarks Numéro de notice : 21988 Affiliation des auteurs : non IGN Thématique : POSITIONNEMENT Nature : Thèse étrangère Note de thèse : Thèse Doctorat : : EPFL : 2018 nature-HAL : Thèse DOI : 10.5075/epfl-thesis-8494 En ligne : https://www.sgc.ethz.ch/publications.html Format de la ressource électronique : URL Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=91986

Réservation

Réserver ce document

Exemplaires (1)

Code-barres	Cote	Support	Localisation	Section	Disponibilité
21988-01	30.70	Livre	Centre de documentation	Géodésie	Disponible

Vision-based detection of aircrafts and UAVs / Artem Rozantsev (2017)  

Permalink

Saisie de la couverture neigeuse de sites avalancheux par des systèmes aéroportés / Julien Vallet (2002)    

Permalink

Nouveaux cadres de référence et transformations de coordonnées en géomatique / François Golay (2001)

Permalink

Proceedings, International workshop on emerging technologies for geo-based applications, Ascona, Switzerland, May 22-25, 2000 / Ecole polytechnique fédérale de Lausanne (2000)

Permalink

Approche de la concertation territoriale par l'analyse lexicale systémique et l'analyse lexicale du discours des acteurs / Wladimir Major (1999)

Permalink

Concept et méthode de gestion des eaux en milieu semi-aride à l'aide d'un système d'information géographique / M. Nguetora (1999)

Permalink

Robustesse dans l'estimation et l'ajustement du variogramme en géostatistique / Marc G. Genton (1996)

Permalink

Cours sur la méthode GPS, 26-28 mars 1990 / H. Dupraz (1990)

Permalink

Proceedings of the 9th international conference on Entity-Relationship Approach, 8-10 October, 1990, Lausanne, Switzerland / Hannu Kangassalo (1990)  

Permalink

Analyse et synthèse d'images de textures structurées / Patrick Volet (1987)

Permalink