Real-time video-plus-depth content creation utilizing time-of-flight sensor - from capture to display
Chuchvara, Aleksandra (2014)
2014
Master's Degree Programme in Information Technology
Tieto- ja sähkötekniikan tiedekunta - Faculty of Computing and Electrical Engineering
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2014-10-08
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tty-201409251444
https://urn.fi/URN:NBN:fi:tty-201409251444
Tiivistelmä
Recent developments in 3D camera technologies, display technologies and other related fields have been aiming to provide 3D experience for home user and establish services such as Three-Dimensional Television (3DTV) and Free-Viewpoint Television (FTV). Emerging multiview autostereoscopic displays do not require any eyewear and can be watched by multiple users at the same time, thus are very attractive for home environment usage. To provide a natural 3D impression, autostereoscopic 3D displays have been design to synthesize multi-perspective virtual views of a scene using Depth-Image-Based Rendering (DIBR) techniques. One key issue of DIBR is that scene depth information in a form of a depth map is required in order to synthesize virtual views. Acquiring this information is quite complex and challenging task and still an active research topic.
In this thesis, the problem of dynamic 3D video content creation of real-world visual scenes is addressed. The work assumed data acquisition setting including Time-of-Flight (ToF) depth sensor and a single conventional video camera. The main objective of the work is to develop efficient algorithms for the stages of synchronous data acquisition, color and ToF data fusion, and final view-plus-depth frame formatting and rendering.
The outcome of this thesis is a prototype 3DTV system capable for rendering live 3D video on a 3D autostereoscopic display. The presented system makes extensive use of the processing capabilities of modern Graphics Processing Units (GPUs) in order to achieve real-time processing rates while providing an acceptable visual quality. Furthermore, the issue of arbitrary view synthesis is investigated in the context of DIBR and a novel approach based on depth layering is proposed. The proposed approach is applicable for general virtual views synthesis, i.e. in terms of different camera parameters such as position, orientation, focal length and varying sensors spatial resolutions. The experimental results demonstrate real-time capability of the proposed method even for CPU-based implementations. It compares favorably to other view synthesis methods in terms of visual quality, while being more computationally efficient.
In this thesis, the problem of dynamic 3D video content creation of real-world visual scenes is addressed. The work assumed data acquisition setting including Time-of-Flight (ToF) depth sensor and a single conventional video camera. The main objective of the work is to develop efficient algorithms for the stages of synchronous data acquisition, color and ToF data fusion, and final view-plus-depth frame formatting and rendering.
The outcome of this thesis is a prototype 3DTV system capable for rendering live 3D video on a 3D autostereoscopic display. The presented system makes extensive use of the processing capabilities of modern Graphics Processing Units (GPUs) in order to achieve real-time processing rates while providing an acceptable visual quality. Furthermore, the issue of arbitrary view synthesis is investigated in the context of DIBR and a novel approach based on depth layering is proposed. The proposed approach is applicable for general virtual views synthesis, i.e. in terms of different camera parameters such as position, orientation, focal length and varying sensors spatial resolutions. The experimental results demonstrate real-time capability of the proposed method even for CPU-based implementations. It compares favorably to other view synthesis methods in terms of visual quality, while being more computationally efficient.