Scalable rendering of massive triangle meshes on light field displays

Abstract

We report on a multiresolution rendering system driving light field displays based on a specially arranged array of projectors and a holographic screen. The system gives multiple freely moving naked-eye viewers the illusion of seeing and manipulating 3D objects with continuous viewer-independent parallax. Multi-resolution techniques which take into account the displayed light field geometry are employed to dynamically adapt model resolution to display capabilities and timing constraints. The approach is demonstrated on two different scales: a desktop PC driving a 7.4Mbeams TV-size display, and a cluster-parallel solution driving a large (1.6x0.9 meters) 35Mbeams display which supports a room-size working space. In both cases, massive meshes of tens of millions of triangles are manipulated at interactive rates.

An interactive 3D medical visualization system based on a light field display

“…Moreover, the optical characteristics of these displays impose specialized rendering methods. In previous work, we have shown that displays of this class can be used for complex data [3,1] by developing customized hardware accelerated rendering techniques. In order to obtain all of the other features required for a prototype 3D medical workstation, we integrated our methods into a DICOM environment, added a simple interaction system based on a 3D mouse, an optimized screen, virtual object and user positions in order to obtain the best perceived quality and the easiest multiuser interaction.…”

Fox

“…For cluster-parallel rendering, we use a sort-first parallel rendering approach with an adaptive out-of-core GPU renderer per back-end node, rather than using an object-based server-push philosophy as in previous light field display rendering systems [Bettio et al 2008]. This makes it possible to reduce server burden and supports the use of different refinement strategies for the rendering and control system.…”

“…Each back-end process controls a frame-buffer portion where it renders the multiresolution model, adaptively loaded from out-of-core, and some visual feedback for the motion control. Differently from previous light field display rendering systems [Bettio et al 2008], we do not push data to rendering nodes from the front end, but let each back-end node manage an adaptively refined version of the model. A multi-pass rendering approach is used, in which a first geometry pass uses vertex shaders that implement the display specific projection, and a series of full-image passes implemented by fragment shaders realize deferred shading and filter the image to produce the required visual effects.…”

“…Such devices give multiple freely moving nakedeye viewers the illusion of seeing and manipulating 3D objects with continuous horizontal parallax. While previous work has demonstrated the possibility of rendering life-like massive models on such displays [Bettio et al 2008], this display technology raises specific user interface issues which have been, so far, neglected.…”