Centre de documentation
scientifique

Real-time multimodal semantic scene understanding for autonomous UGV navigation / Yifei Zhang (2021)  

Public Titre : Real-time multimodal semantic scene understanding for autonomous UGV navigation Type de document : Thèse/HDR Auteurs : Yifei Zhang, Auteur ; Fabrice Mériaudeau, Directeur de thèse ; Désiré Sidibé, Directeur de thèse Editeur : Dijon : Université Bourgogne Franche-Comté UBFC Année de publication : 2021 Importance : 114 p. Format : 21 x 30 cm Note générale : Bibliographie Thèse pour obtenir le doctorat de l'Université Bourgogne Franche-Comté, Spécialité Instrumentation et informatique d’image Langues : Anglais (eng) Descripteur : [Vedettes matières IGN] Traitement d'image mixte [Termes IGN] données polarimétriques [Termes IGN] fusion d'images [Termes IGN] image RVB [Termes IGN] intégration de données [Termes IGN] navigation autonome [Termes IGN] segmentation sémantique [Termes IGN] temps réel [Termes IGN] véhicule sans pilote Index. décimale : THESE Thèses et HDR Résumé : (Auteur) Robust semantic scene understanding is challenging due to complex object types, as well as environmental changes caused by varying illumination and weather conditions. This thesis studies the problem of deep semantic segmentation with multimodal image inputs. Multimodal images captured from various sensory modalities provide complementary information for complete scene understanding. We provided effective solutions for fully-supervised multimodal image segmentation and few-shot semantic segmentation of the outdoor road scene. Regarding the former case, we proposed a multi-level fusion network to integrate RGB and polarimetric images. A central fusion framework was also introduced to adaptively learn the joint representations of modality-specific features and reduce model uncertainty via statistical post-processing.In the case of semi-supervised semantic scene understanding, we first proposed a novel few-shot segmentation method based on the prototypical network, which employs multiscale feature enhancement and the attention mechanism. Then we extended the RGB-centric algorithms to take advantage of supplementary depth cues. Comprehensive empirical evaluations on different benchmark datasets demonstrate that all the proposed algorithms achieve superior performance in terms of accuracy as well as demonstrating the effectiveness of complementary modalities for outdoor scene understanding for autonomous navigation. Note de contenu : 1. Introduction 1.1 Context and Motivation 1.2 Background and Challenges 1.3 Contributions 1.4 Organization 2. Background on Neural Networks 2.1 Basic Concepts 2.2 Neural Network Layers 2.3 Optimization 2.4 Model Training 2.5 Evaluation Metrics 2.6 Summary 3. Literature Review 3.1 Fully-supervised Semantic Image 3.2 Datasets 3.3 Summary 4. Deep Multimodal Fusion for Semantic Image Segmentation 4.1 CMNet: Deep Multimodal Fusion 4.2 A Central Multimodal Fusion Framework 4.3 Summary 5. Few-shot Semantic Image Segmentation 5.1 Introduction on Few-shot Segmentation 5.2 MAPnet: A Multiscale Attention-Based Prototypical Network 5.3 RDNet: Incorporating Depth Information into Few-shot Segmentation 5.4 Summary 6. Conclusion and Future Work 6.1 General Conclusion 6.2 Future Perspectives Numéro de notice : 26527 Affiliation des auteurs : non IGN Thématique : IMAGERIE Nature : Thèse française Note de thèse : Thèse de Doctorat : Instrumentation et informatique d’image : Bourgogne : 2021 nature-HAL : Thèse Date de publication en ligne : 02/03/2021 En ligne : https://hal.science/tel-03154783v1 Format de la ressource électronique : URL Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=97556