التفاصيل البيبلوغرافية
| العنوان: |
Towards Robust Visual SLAM in Low-Textured Environments |
| المؤلفون: |
Company Córcoles, Juan José |
| المساهمون: |
Ortiz Rodríguez, Alberto, García Fidalgo, Emilio |
| بيانات النشر: |
Universitat de les Illes Balears |
| سنة النشر: |
2023 |
| المجموعة: |
Universitat de les Illes Balears: DSpace |
| مصطلحات موضوعية: |
SLAM, Visual odometry, Loop-closure detection, Computer vision, 004 - Informàtica, 621.3 - Enginyeria elèctrica. Electrotècnia. Telecomunicacions, L7 Robótica industrial, móvil y de servicios |
| الوصف: |
[eng] Localization and mapping are two fundamental tasks in several domains, such as mobile robotics or augmented/virtual reality. These tasks involve, respectively, to accurately determine the position and orientation of an agent within its operating environment and construct a map of its surroundings. Given their close relationship, Simultaneous Localization and Mapping (SLAM) algorithms propose to solve these tasks at the same time in an incremental manner. SLAM has been carried out using different sensors for years. When cameras are used as the main source of information, these approaches are known as Visual SLAM. While traditional Visual SLAM algorithms have achieved a high level of accuracy in textured environments, they may struggle in low-textured environments due to a lack of visual features on them. This can lead to inaccurate motion estimates, a degraded map quality or even failures. Under this context, this thesis address the challenge of improving the accuracy, reliability and robustness of Visual SLAM systems in low-textured environments. With this objective in mind, we propose two novel Visual Odometry (VO) and two Loop Closure Detection (LCD) methods that combine traditional approaches with geometrical features of the scene, such as lines or planes, typically more present in low-textured scenarios. The accuracy, adaptability and effectiveness of the proposed methods are thoroughly validated through a comprehensive set of experiments. The results show that our solutions present a better accuracy and performance, in contrast to other state-of-the-art methods, in low-textured environments and, additionally, perform favourably in other scenarios, demonstrating the potential of incorporating geometrical features into visual SLAM algorithms in these cases. ; [spa] La localización y el mapeo son dos tareas fundamentales en diversos ámbitos, tales como la robótica móvil o la realidad aumentada/virtual. Estas tareas implican, respectivamente, determinar con precisión la posición y orientación de un agente ... |
| نوع الوثيقة: |
doctoral or postdoctoral thesis |
| وصف الملف: |
application/pdf |
| اللغة: |
English |
| Relation: |
http://hdl.handle.net/11201/160972 |
| الاتاحة: |
http://hdl.handle.net/11201/160972 |
| Rights: |
all rights reserved ; info:eu-repo/semantics/openAccess |
| رقم الانضمام: |
edsbas.CE365580 |
| قاعدة البيانات: |
BASE |