Advancing 3D Scene Reconstruction: Techniques, Pipelines, and Applications
(2024) MAMM15 20241Department of Design Sciences
- Abstract
- This master thesis is carried out at Sony Nordic which aims to investigate state-of-the-art methods on 3D Scene reconstruction and explores the potential utilization in the industry. The project addresses challenges in reconstructing complex scenes using both static camera setups which consider scenarios with freely moving rigid objects and freely moving cameras. Throughout the research, several key questions were answered, resulting in a robust pipeline including image capture, camera calibration, foreground segmentation, camera estimation, and model training. The reconstruction utilizes technologies such as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting. The result of this work demonstrates the feasibility and highlights the... (More)
- This master thesis is carried out at Sony Nordic which aims to investigate state-of-the-art methods on 3D Scene reconstruction and explores the potential utilization in the industry. The project addresses challenges in reconstructing complex scenes using both static camera setups which consider scenarios with freely moving rigid objects and freely moving cameras. Throughout the research, several key questions were answered, resulting in a robust pipeline including image capture, camera calibration, foreground segmentation, camera estimation, and model training. The reconstruction utilizes technologies such as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting. The result of this work demonstrates the feasibility and highlights the potential challenges of 3D reconstruction under various camera settings. Additionally, we propose several applications that could benefit from these advancements, depending on the scenarios. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9162342
- author
- Tian, Xuening LU
- supervisor
- organization
- course
- MAMM15 20241
- year
- 2024
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- 3D reconstruction, Computer Vision, Neural Radiance Fields, Gaussian Splatting, Rendering, Deep Learning
- language
- English
- id
- 9162342
- date added to LUP
- 2024-06-13 08:11:25
- date last changed
- 2024-06-13 08:11:25
@misc{9162342, abstract = {{This master thesis is carried out at Sony Nordic which aims to investigate state-of-the-art methods on 3D Scene reconstruction and explores the potential utilization in the industry. The project addresses challenges in reconstructing complex scenes using both static camera setups which consider scenarios with freely moving rigid objects and freely moving cameras. Throughout the research, several key questions were answered, resulting in a robust pipeline including image capture, camera calibration, foreground segmentation, camera estimation, and model training. The reconstruction utilizes technologies such as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting. The result of this work demonstrates the feasibility and highlights the potential challenges of 3D reconstruction under various camera settings. Additionally, we propose several applications that could benefit from these advancements, depending on the scenarios.}}, author = {{Tian, Xuening}}, language = {{eng}}, note = {{Student Paper}}, title = {{Advancing 3D Scene Reconstruction: Techniques, Pipelines, and Applications}}, year = {{2024}}, }