Approximating ray traced reflections using screenspace data

Abstract

At Spark Vision, independent component images, each containing a part of the rendered geometry, are layered to create complete images. Due to the assumption of independence; reflections cannot be accurately rendered. A screen-space method for adding reflections to a rendered image using buffers for geometry and surface properties is proposed. Reflections are traced using an approximation of ray tracing. The goal is to allow for the continued use of component images with reflections added as a post-processing effect in real time. The method outlined allows for perfect and glossy reflections comparable to the quality of commercial ray tracers for optimal scenes. However, it fails to capture reflections of objects that are not visible from the camera view-point except for cases in which these are part of a pre-defined surrounding environment. The method itself allows for dynamic scenes, camera settings and surface properties including BRDFs if used as an off-line renderer. Due to demands of image quality, time constraints and the lack of GPU support a method for caching the rays is proposed based on assumption of a static geometry. The caching method allows for arbitrary variations in lighting, textures and reflection strength as well as limited variations of normal mapping and surface shininess. This method achieves acceptable running times for the intended application.