Impostor-based Rendering Acceleration for Virtual, Augmented, and Mixed Reality
Martin Misiak, Arnulph Fuhrmann, and Marc Erich Latoschik
VRST 2021
Session: Paper 1: Tracking, Rendering and Social Interaction
Abstract
This paper presents an image-based rendering approach to accelerate rendering time of virtual scenes containing a large number of complex high poly count objects. Our approach replaces complex objects by impostors, light-weight image-based representations leveraging geometry and shading related processing costs. In contrast to their classical implementation, our impostors are specifically designed to work in Virtual-, Augmented- and Mixed Reality scenarios (XR for short), as they support stereoscopic rendering to provide correct depth perception. Motion parallax of typical head movements is compensated by using a ray marched parallax correction step. Our approach provides a dynamic run-time recreation of impostors as necessary for larger changes in view position. The dynamic run-time recreation is decoupled from the actual rendering process. Hence, its associated processing cost is therefore distributed over multiple frames. This avoids any unwanted frame drops or latency spikes even for impostors of objects with complex geometry and many polygons. In addition to the significant performance benefit, our impostors compare favorably against the original mesh representation, as geometric and textural temporal aliasing artifacts are heavily suppressed.
DOI:: [ Ссылка ]
WEB:: [ Ссылка ]
Videos for VRST 2021
