Descripteur
Documents disponibles dans cette catégorie (1332)
![](./images/expand_all.gif)
![](./images/collapse_all.gif)
Etendre la recherche sur niveau(x) vers le bas
Titre : Reshaping perception for autonomous driving with semantic keypoints Type de document : Thèse/HDR Auteurs : Lorenzo Bertoni, Auteur Editeur : Lausanne : Ecole Polytechnique Fédérale de Lausanne EPFL Année de publication : 2022 Importance : 177 p. Format : 21 x 30 cm Note générale : bibliographie
Thèse présentée pour l'obtention du grade de Docteur ès Sciences, Ecole Polytechnique Fédérale de LausanneLangues : Anglais (eng) Descripteur : [Vedettes matières IGN] Intelligence artificielle
[Termes IGN] classification par réseau neuronal convolutif
[Termes IGN] détection automatique
[Termes IGN] détection d'objet
[Termes IGN] détection de piéton
[Termes IGN] estimation de pose
[Termes IGN] navigation autonome
[Termes IGN] système multi-agents
[Termes IGN] vision par ordinateurRésumé : (auteur) The field of artificial intelligence is set to fuel the future of mobility by driving forward the transition from advanced driver-assist systems to fully autonomous vehicles (AV). Yet the current technology, backed by cutting-edge deep learning techniques, still leads to fatal accidents and does not convey trust. Current frameworks for 3D perception tasks, such as 3D object detection, are not adequate as they (i) do not generalize well to new scenarios, (ii) do not take into account measures of confidence in their predictions, and (iii) are not suitable for large-scale deployment as mainly based on costly LiDAR sensors. This doctoral thesis aims to study vision-based deep learning frameworks that can accurately perceive the world in 3D and generalize to new scenarios. We propose to escape the pixel domain using semantic keypoints, a sparse representation for every object in the scene containing meaningful information for 2D and 3D reasoning. The low-dimensionality enables downstream neural networks to focus on essential elements in the scene and improve their generalization capabilities. Furthermore, driven by the limitation of deep learning architectures outputting point estimates, we study how to estimate a confidence interval for each prediction. In particular, we emphasize vulnerable road users, such as pedestrians and cyclists, and explicitly address the long tail of 3D pedestrian detection to contribute to the safety of our roads. We further show the efficacy of our framework on multiple real-world domains by (a) integrating it in an existing AV pipeline, (b) detecting human-robot eye contact in real-world scenarios, and (c) helping verify the compliance of safety measures in the case of the COVID-19 outbreak. Finally, we publicly release the source code of all our projects and develop a unified library to contribute to an open science mission. Note de contenu : 1- Introduction
2- Semantic keypoints detection
3- Monocular 3D pedestrian localization and uncertainty estimation
4- Tackling the long tail of 3D pedestrian localization with stereo cameras
5- Autonomous driving applications of pedestrian 3D detection
6- Detecting pedestrians attention: Human-robot eye contact in the wild
7- Beyond autonomous driving: Social interactions and social distancing
8- ConclusionNuméro de notice : 24077 Affiliation des auteurs : non IGN Thématique : INFORMATIQUE Nature : Thèse étrangère Note de thèse : PhD Thesis : Sciences : EPFL : 2022 DOI : 10.5075/epfl-thesis-10072 En ligne : https://doi.org/10.5075/epfl-thesis-10072 Format de la ressource électronique : URL Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=102212
Titre : Robustness of visual SLAM techniques to light changing conditions : Influence of contrasted local features, multi-planar representations and multimodal image analysis Type de document : Thèse/HDR Auteurs : Xi Wang, Auteur ; Eric Marchand, Directeur de thèse Editeur : Rennes : Université de Rennes 1 Année de publication : 2022 Importance : 153 p. Format : 21 x 30 cm Note générale : bibliographie
Thèse de Doctorat de l'Université de Rennes 1, Spécialité InformatiqueLangues : Anglais (eng) Descripteur : [Vedettes matières IGN] Traitement d'image optique
[Termes IGN] apprentissage profond
[Termes IGN] cartographie et localisation simultanées
[Termes IGN] éclairage
[Termes IGN] estimation de pose
[Termes IGN] information sémantique
[Termes IGN] primitive géométrique
[Termes IGN] programmation linéaire
[Termes IGN] robotique
[Termes IGN] vision par ordinateurIndex. décimale : THESE Thèses et HDR Résumé : (auteur) The SLAM (Simultaneous Localization And Mapping) technique concentrates on localizing and recovering the environment in a simultaneous way and is one of the core functionalities of many industrial products such as augmented reality, where the device poses should be tracked in real-time; autonomous driving, where one needs to localize the vehicle in a pre-generated map or unknown environment; and even modern filmmaking workflow, where the relative camera position and orientation are critical for post-processing or real-time prevising for directors and actors to visualise the visual effects on the stage. Multiple difficulties in different levels can influence the final performance of robot agents’s SLAM task, as the pipeline is long and complicated from the real world physics to the required information such as agent poses and 3-D map, which help us visualize colourful graphics scenes in AR devices or make hard decisions on the highway for autonomous driving. Many solutions are proposed for addressing each problem, respectively, with the means from classic statistic probability models to the modern data-driven deep neural network. However, the quest of improving the robot’s robustness under dynamic and complicated environments perisists and becomes more and more significant and active for nowadays robotics research. The need for improving the robustness of robot agents is imminent and regarded as one of most imperative factors for deploying robots ubiquitously in our daily life. Under this context, this thesis tries to address a small drop in the ocean of the problem of SLAM robustness, yet in a very systematic view: we try to break down the SLAM system into different and inter-influential modules. Then use the concept of "divide and conquer" for answering possible questions within each module and wishing to contribute to the community and help improve the robustness of SLAM systems under complicated conditions. With the above objectives, the contributions of the thesis are stated as follows for tackling the robustness problem from multiple angles: 1) From the image feature angle, we proposed a multiple layered image structure for improving the performance of traditional local image features under extreme conditions. Furthermore, an optimization method on linear searching and mutual information assisted convex optimization are designed for tuning the optimal parameters with the proposed structure; 2) From the geometric primitive angle, we proposed a relative pose estimation and SLAM framework under the multiple planar assumption, by keypoint feature-based and template tracker based methods, respectively. We tried to achieve better performance of mapping and tracking simultaneously with the help of a more general planar assumption. 3) From the angle of relocalization of the SLAM system, the idea is to recover the already passed locations of the robot agent for lowering the overall estimation error or when the robot is in lost status. We proposed a binary graph structure for embedding spatial information and heterogeneous data formats such as depth image, semantic information etc. The proposed method enables robotics SLAM systems to relocalize themselves with a higher success rate even under different lighting, weather and seasonal conditions. Note de contenu : 1- Introduction
2- Résumé
3- Background on visual SLAM techniques
4- Related work
5- Organisation
6- Multiple layers image
7- Multi-planar relative pose estimation via superpixel
8- TT-SLAM
9- Binary graph descriptor for robust relocalization on heterogeneous data
ConclusionNuméro de notice : 24074 Affiliation des auteurs : non IGN Thématique : IMAGERIE/INFORMATIQUE Nature : Thèse française Note de thèse : Thèse de Doctorat : Informatique : Rennes 1 : 2022 Organisme de stage : IRISA DOI : sans En ligne : https://www.theses.fr/2022REN1S022 Format de la ressource électronique : URL Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=102162 Scaling up and evaluating surface reconstruction from point clouds of open scenes / Yanis Marchand (2022)
![]()
Titre : Scaling up and evaluating surface reconstruction from point clouds of open scenes Titre original : Passage à l'échelle et évaluation de la reconstruction de surface à partir de nuage de points de scènes ouvertes Type de document : Thèse/HDR Auteurs : Yanis Marchand , Auteur ; Bruno Vallet
, Directeur de thèse ; Laurent Caraffa
, Encadrant
Editeur : Champs-sur-Marne [France] : Université Gustave Eiffel Année de publication : 2022 Note générale : bibliographie
Thèse de doctorat de l’Université Gustave Eiffel, Ecole doctorale n° 532, Math ́ematiques et Sciences et Technologies de l’Information et de la Communication (MSTIC), Spécialité de doctorat : Informatique - Unité de recherche : LASTIG (IGN)Langues : Français (fre) Descripteur : [Vedettes matières IGN] Lasergrammétrie
[Termes IGN] informatique
[Termes IGN] reconstruction d'objet
[Termes IGN] scène 3D
[Termes IGN] semis de points
[Termes IGN] traitement réparti
[Termes IGN] vision par ordinateurIndex. décimale : THESE Thèses et HDR Résumé : (auteur) Cette thèse de doctorat traite de deux aspects de la reconstruction de surface à partir de nuage de points. Premièrement, elle aborde le cas large échelle où un nuage de points est trop volumineux pour être stocké dans la mémoire d'un seul ordinateur. Nous présentons un algorithme distribué de bout en bout permettant de traiter des nuages de points arbitrairement grands tout en garantissant l'étanchéité de la surface produite. Deuxièmement, cette thèse contribue à l'évaluation de la reconstruction de surface de par la définition de deux protocoles. Le premier nécessite des données synthétiques alors que le deuxième peut être mis en place en ayant uniquement recours à des données provenant de capteurs. Ces protocoles et les nouvelles métriques qui leur sont associées permettent de quantifier la qualité des reconstructions avec un biais moins important que les approches utilisées jusqu'alors. Numéro de notice : 17739 Affiliation des auteurs : UGE-LASTIG (2020- ) Thématique : IMAGERIE/INFORMATIQUE Nature : Thèse française Note de thèse : Thèse : Informatique : Gustave Eiffel : 2022 Organisme de stage : LASTIG (IGN) nature-HAL : Thèse DOI : sans En ligne : https://tel.hal.science/tel-04031734 Format de la ressource électronique : URL Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=102877
Titre : Scene understanding and gesture recognition for human-machine interaction Type de document : Thèse/HDR Auteurs : Naina Dhingra, Auteur Editeur : Zurich : Eidgenossische Technische Hochschule ETH - Ecole Polytechnique Fédérale de Zurich EPFZ Année de publication : 2022 Note générale : Bibliographie
A dissertation submitted to attain the degree of Doctor of Sciences of ETH ZurichLangues : Français (fre) Descripteur : [Vedettes matières IGN] Intelligence artificielle
[Termes IGN] apprentissage profond
[Termes IGN] attention (apprentissage automatique)
[Termes IGN] classification orientée objet
[Termes IGN] classification par réseau neuronal convolutif
[Termes IGN] classification par séparateurs à vaste marge
[Termes IGN] compréhension de l'image
[Termes IGN] image RVB
[Termes IGN] interaction homme-machine
[Termes IGN] oculométrie
[Termes IGN] reconnaissance automatique
[Termes IGN] reconnaissance de formes
[Termes IGN] reconnaissance de gestes
[Termes IGN] réseau neuronal récurrent
[Termes IGN] scène
[Termes IGN] vision par ordinateurRésumé : (auteur) Scene understanding and gesture recognition are useful for a myriad of applications such as human-robotic interaction, assisting blind and visually impaired people, advanced driver assistance systems, and autonomous driving. To work autonomously in real-world environments, automatic systems need to deliver non-verbal information to enhance the verbal communication in particular for blind people. We are exploring the holistic approach for providing the scene as well as gesture related information. We propose that incorporating attention mechanisms in neural networks which behave similarly to attention in the human brain, and conducting an integrated study using neural networks in real-time can yield significant improvements in the scene and gesture understanding, thereby enhancing the user experience. In this thesis, we investigate the understanding of visual scenes and gestures. We explore these two areas, in particular, by proposing novel architectures, training methods, user studies, and thorough evaluations. We show that, for deep learning approaches, attention or self attention mechanisms improve and push the boundaries of network performance for different tasks in consideration. We suggest that the various kinds of gestures can complement and supplement each other’s information to better understand non-verbal conversation; hence integrated gestures comprehension is useful. First, we focus on visual scene understanding using scene graph generation. We propose, BGT-Net, a new network that uses an object detection model with 1) bidirectional gated recurrent units for object-object communication and 2) transformer encoders including self attention to classify the objects and their relationships. We address the problem of bias caused by the long tailed distribution in the dataset. This enables the network to perform even for the unseen objects or relationships in the dataset. Second, we propose to learn hand gesture recognition from RGB and RGB-D videos using attention learning. We present a novel architecture based on residual connections and an attention mechanism. Our approach successfully detects hand gestures when evaluated on three open-source datasets. Third, we explore pointing gesture recognition and localization using open-source software, i.e. OpenPtrack which uses a deep learning based iii network to track multi-persons in the scene. We use a Kinect sensor as an input device and conduct a user study with 26 users to evaluate the system using two setup types. Fourth, we propose a technique to perform eye gaze tracking using OpenFace which is based on a deep learning model and RGB webcam. We use support vector machine regression to estimate the position of eye gaze on the screen. In a study, we evaluate the system with 28 users and show that this system can perform similarly to commercially expensive eye trackers. Finally, we focus on 3D head pose estimation using two models: 1)headPosr includes residual connections for the base network followed by a transformer encoder. It outperforms existing models but has a drawback of being computationally expensive; 2) lwPosr uses depthwise separable convolutions and transformer encoders. It is a two stream network in fine-grained fashion to estimate the three angles of the head pose. We demonstrate that this method is able to predict head poses better than state-of-the-art lightweight networks. Note de contenu : 1- Introduction
2- Background
3- State of the art
4- Scene graph generation
5- 3D hand gesture recognition
6- Pointing gesture recognition
7- Eye-gaze tracking
8- Head pose estimation
9- Lightweight head pose estimation
10- SummaryNuméro de notice : 24039 Affiliation des auteurs : non IGN Thématique : IMAGERIE/INFORMATIQUE Nature : Thèse étrangère Note de thèse : PhD Thesis : Sciences : ETH Zurich :2022 DOI : sans En ligne : https://www.research-collection.ethz.ch/handle/20.500.11850/559347 Format de la ressource électronique : URL Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=101876
Titre : Schematizing crossroads from abstract textual descriptions Type de document : Article/Communication Auteurs : Jean-Marie Favreau, Auteur ; Guillaume Touya , Auteur ; Jérémy Kalsron, Auteur
Editeur : Bonn : Université de Bonn Année de publication : 2022 Projets : ACTIVmap / Favreau, Jean-Marie Conférence : CompCarto 2022, 1st workshop on Computational Cartography 19/05/2022 20/05/2022 Bonn Allemagne programme Importance : 3 p. Note générale : Bibliographie Langues : Anglais (eng) Descripteur : [Termes IGN] carrefour
[Termes IGN] carte tactile
[Termes IGN] cartogramme
[Termes IGN] exploration de texte
[Vedettes matières IGN] CartologieRésumé : (auteur) [début] The use of cartographic representations among people with visual impairments (PVI) is often limited by the lack of available materials. However, two uses have been identified: diagrams made with sticks magnetised to a metal plate (Figure 1) are used by Orientation and Mobility instructors as a discussion aid around complex areas (typically intersections), and more accurate maps made by transcribing adapters are sometimes produced for regular use. While classical variations of the generalisation and stylisation approaches allow for the production of fairly accurate maps [JLCJ21], for example from OpenStreetMap data (figure 2), there are currently no known approaches to producing a more schematic representation, in the manner of the locomotion instructors’ magnets. Numéro de notice : C2022-013 Affiliation des auteurs : UGE-LASTIG+Ext (2020- ) Thématique : GEOMATIQUE Nature : Communication nature-HAL : ComSansActesPubliés-Unpublished DOI : sans Date de publication en ligne : 24/05/2022 En ligne : https://hal.science/hal-03677334/ Format de la ressource électronique : URL Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=100747 Self-attention and generative adversarial networks for algae monitoring / Nhut Hai Huynh in European journal of remote sensing, vol 55 n° 1 (2022)
PermalinkSemantic segmentation of high-resolution remote sensing images based on a class feature attention mechanism fused with Deeplabv3+ / Zhimin Wang in Computers & geosciences, vol 158 (January 2022)
PermalinkSimulation of dispersion effects by considering interactions of pedestrians and bicyclists using an agent space model / Mingwei Liu in Computers, Environment and Urban Systems, vol 91 (January 2022)
PermalinkPermalinkStudying informativeness of satellite image texture for sea ice state retrieval using deep learning methods / Clément Fougerouse (2022)
PermalinkTowards expressive graph neural networks : Theory, algorithms, and applications / Georgios Dasoulas (2022)
PermalinkTowards synthetic sensing for smart cities : a machine/deep learning-based approach / Faraz Malik Awan (2022)
PermalinkUnsupervised generative models for data analysis and explainable artificial intelligence / Mohanad Abukmeil (2022)
PermalinkUrban infrastructure audit: an effective protocol to digitize signalized intersections by mining street view images / Xiao Li in Cartography and Geographic Information Science, vol 49 n° 1 (January 2022)
PermalinkPermalink