Improve depth estimation based on deep learning and information fusion
Depth estimation is a highly focused research direction in the field of computer vision, and it has seen rapid development and a wealth of research results in recent years. However, current mainstream depth estimation technologies rely on computationally expensive deep learning methods or direct dep...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis-Master by Coursework |
| Language: | English |
| Published: |
Nanyang Technological University
2024
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/173222 |
| _version_ | 1811690816959152128 |
|---|---|
| author | Xue, Mingqing |
| author2 | Mao Kezhi |
| author_facet | Mao Kezhi Xue, Mingqing |
| author_sort | Xue, Mingqing |
| collection | NTU |
| description | Depth estimation is a highly focused research direction in the field of computer vision, and it has seen rapid development and a wealth of research results in recent years. However, current mainstream depth estimation technologies rely on computationally expensive deep learning methods or direct depth acquisition technologies that require costly, specialized sensor equipment, such as RGB-D cameras and LIDAR. These technologies have some practical limitations, such as the need for high computational power and reliance on specialized hardware.
In response to these issues, this study proposes a depth estimation algorithm suitable for deployment in smartphone applications, aiming to achieve fast and accurate monocular depth estimation on low-power devices. We adopted an innovative approach that combines deep learning techniques with classic geometric depth estimation methods (such as SfM), leveraging geometric constraints to reduce computational complexity and runtime. This hybrid approach not only optimizes the efficiency of depth estimation but also maintains the accuracy and robustness of the results.
Through a series of rigorous experimental designs and validations, the research results demonstrate the advantages of the proposed method over traditional algorithms in low-power environments. This research not only provides a new academic perspective but also has broad application prospects in practical applications, especially in the field of mobile device applications. With the continuous improvement of smartphone processing capabilities and further optimization of deep learning technologies, it is expected that the method proposed by this study will provide a new solution for mobile visual applications, pushing the ability of smartphones in three-dimensional space perception to a new height. |
| first_indexed | 2024-10-01T06:10:01Z |
| format | Thesis-Master by Coursework |
| id | ntu-10356/173222 |
| institution | Nanyang Technological University |
| language | English |
| last_indexed | 2024-10-01T06:10:01Z |
| publishDate | 2024 |
| publisher | Nanyang Technological University |
| record_format | dspace |
| spelling | ntu-10356/1732222024-01-19T15:44:56Z Improve depth estimation based on deep learning and information fusion Xue, Mingqing Mao Kezhi School of Electrical and Electronic Engineering EKZMao@ntu.edu.sg Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision Depth estimation is a highly focused research direction in the field of computer vision, and it has seen rapid development and a wealth of research results in recent years. However, current mainstream depth estimation technologies rely on computationally expensive deep learning methods or direct depth acquisition technologies that require costly, specialized sensor equipment, such as RGB-D cameras and LIDAR. These technologies have some practical limitations, such as the need for high computational power and reliance on specialized hardware. In response to these issues, this study proposes a depth estimation algorithm suitable for deployment in smartphone applications, aiming to achieve fast and accurate monocular depth estimation on low-power devices. We adopted an innovative approach that combines deep learning techniques with classic geometric depth estimation methods (such as SfM), leveraging geometric constraints to reduce computational complexity and runtime. This hybrid approach not only optimizes the efficiency of depth estimation but also maintains the accuracy and robustness of the results. Through a series of rigorous experimental designs and validations, the research results demonstrate the advantages of the proposed method over traditional algorithms in low-power environments. This research not only provides a new academic perspective but also has broad application prospects in practical applications, especially in the field of mobile device applications. With the continuous improvement of smartphone processing capabilities and further optimization of deep learning technologies, it is expected that the method proposed by this study will provide a new solution for mobile visual applications, pushing the ability of smartphones in three-dimensional space perception to a new height. Master's degree 2024-01-19T01:21:57Z 2024-01-19T01:21:57Z 2023 Thesis-Master by Coursework Xue, M. (2023). Improve depth estimation based on deep learning and information fusion. Master's thesis, Nanyang Technological University, Singapore. https://hdl.handle.net/10356/173222 https://hdl.handle.net/10356/173222 en application/pdf Nanyang Technological University |
| spellingShingle | Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision Xue, Mingqing Improve depth estimation based on deep learning and information fusion |
| title | Improve depth estimation based on deep learning and information fusion |
| title_full | Improve depth estimation based on deep learning and information fusion |
| title_fullStr | Improve depth estimation based on deep learning and information fusion |
| title_full_unstemmed | Improve depth estimation based on deep learning and information fusion |
| title_short | Improve depth estimation based on deep learning and information fusion |
| title_sort | improve depth estimation based on deep learning and information fusion |
| topic | Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision |
| url | https://hdl.handle.net/10356/173222 |
| work_keys_str_mv | AT xuemingqing improvedepthestimationbasedondeeplearningandinformationfusion |