Light-field Display Architecture and the Challenge of Synthetic Light-field Radiance Image Rendering

摘要

The ability to render 3D data/scenes onto 2D displays have greatly enhanced the human ability to visualize complex scenes. However, when 3D information is projected in 2D, many of the natural depth cues are lost or are incorrect requiring a higher cognitive effort from the viewer to reconstruct the scene. A light-field display projects perspective correct, full parallax 3D aerial imagery independent of the number of viewers or viewer positions. As such, light-field displays enable collaboration and promote the analysis of complex information; however, the generation of the synthetic light-field is computationally challenging with off-the-shelf GPUs. Light-field projection fidelity aside, the high Size, Weight and Power (SWaP) cost of light-field rendering ultimately limits the deployment of light-field display systems. This is partly due to the issue that the modern GPUs and rendering APIs do not support multi-view rendering natively.