面向裸眼三维显示的光场数据分析与高效绘制技术

Recent years, the field of three-dimensional display has become one with a fascinating variation and depth of research. Many devices, techniques and methods have been proposed to provide human natural 3D experience beyond conventional 2D displays. Among these approaches, the automultiscopic display technique is one of the most important techniques. This is mainly because it does not have special requirements on observers and produces images that exhibit all the characteristics of the original scene. To achieve this, a large number of views with realistic images or tons of rays of the light field of objects are produced in order to give the high fidelity of stereo and the appearance of smooth motion parallax. However, current graphics hardware and rendering pipeline are designed for single view graphics. Rendering a large amount of rays or many images for arbitrary views brings challenges for computer graphics. In this proposal, we study the theorical and practical problems of efficient and fast light field synthesis for automultiscopic display. At first, we plan to conduct user studies on the perceptual 3D cues of human vision, so as to have better understand of how the distribution of rays in a light field really produce 3D cues for human. A representation of light field with perceptual criteria will be formulated and proposed. Then, we take spectral analysis for light field data. New parameterizations that are applicable for automultiscopic display, the band-limited signal of light transports and other problems of light field will be analyzed in signal-processing framework. Based on these two researches, we will develop new light field synthesis and rendering algorithms for real-time applications on automultiscopic displays. Finally, we will validate and demonstrate our theories, analysis, algorithms and implementations on certain automultiscopic displays. Our goal is to derive a theorical framework, analysis and implementations for all mechanisms of efficient and fast light field synthesis and rendering, which will bring new concept and benefit future researches on automultiscopic display techniques.

近年来三维显示技术获得了长足的进步，能够提供自然、真实三维显示环境的裸眼三维显示技术成为重要的发展方向。裸眼三维显示技术为了实现对三维对象虚拟光场的逼真模拟，需要计算机能够支持不同空间位置、不同方向观看的具有高真实感的三维内容的呈现。这对当前的计算机绘制技术提出了新的挑战。.本项目研究着眼于面向裸眼三维显示的光场数据高效绘制所需要的理论与实践问题，计划以裸眼三维显示所需的大规模、高真实感光场数据为研究目标，重点开展基于视觉感知特征的光场数据的理论表示，光场数据的信号分析研究，并结合图形学研究的最新进展突破光场数据的高效绘制技术。通过构建裸眼三维显示绘制引擎，实现对光场数据分析与高效绘制技术的验证。通过本项目的研究，将为面向裸眼三维显示的光场数据的生成提供理论与分析的依据，并为实现高真实感、高逼真度的裸眼三维显示技术的研发提供理论参考与数据生成技术的支撑。

近年来三维显示技术获得了长足的进步，能够提供自然、真实三维显示环境的裸眼三维显示技术成为重要的发展方向。在这为期一年的项目研究中，我们着眼于研究面向裸眼三维显示的绘制与显示技术并搭建硬件平台进行验证：提出了针对大规模场景的真实感绘制方法、针对动态场景的全频光照实时真实感绘制方法；研发了可拓展式光场三维显示装置与自动校正方法，并搭建了远程三维显示交互原型系统。通过本项目的研究，为面向裸眼三维显示的光场数据生成、绘制与显示提供了新的方法，从而为实现高真实感、高逼真度的裸眼三维显示技术的进一步发展提供了理论与技术支持。在本项目支持的一年研究期间共计发表（含录用）论文4篇（SCI 3篇，EI 1篇，包括本领域的国际顶级期刊ACM Trans. on Graphics论文1篇），申请专利2项，搭建原型系统一套。