Real-Time Light Field Video Focusing and GPU Accelerated Streaming

This paper proposes a novel solution of real-time depth range and correct focusing estimation in light field videos represented by arrays of video sequences. This solution, compared to previous approaches, offers a novel way to render high-quality synthetic views from light field videos on contemporary hardware in real-time. Only the video frames containing color information with no other attributes, such as captured depth, are needed. The drawbacks of the previous proposals such as block artifacts in the defocused parts of the scene or manual setting of the depth range are also solved in this paper. This paper describes a complete solution that solves the main memory and performance issues of light field rendering on contemporary personal computers. The whole integration of high-quality light field videos supersedes the approaches in previous works and the paper also provides measurements and experimental results. While reaching the same quality as slower state-of-the-art approaches, this method can still be used in real-time which makes it suitable for industry and real-life scenarios as an alternative to standard 3D rendering approaches.