The exploitation of image construction data and temporal/image coherence in ray traced animation
Dana Marshall, Donald S. Fussell · 2001
The similarity in successive images generated for computer graphic animated sequences is a source of great possibilities for accelerating rendering time by avoiding seemingly redundant calculations. There is a wealth of information that a renderer calculates or accesses in the course of creating an image that can be used to aid the creation of future frames. Previous solutions either incurred visibility errors or used methods to guarantee the accuracy of images that could be prohibitively time consuming. This dissertation proposes an elegant and fast method that speeds the calculation of ray traced animated scenes constructed of convex objects by a combination of temporal and image coherence. The algorithm reuses rays calculated from previous frames by transforming them to their new locations and updating their color from their shading and material properties. The solution presented guarantees the accuracy of the visibility and shading of reused samples while minimizing calculation time. The method works with any existing bounding volume acceleration method and is compatible with jittered anti-aliasing schemes. It handles reflections and refractions and does not depend upon any part of the scene remaining static. It does assume the existence of some continuity in motion, which is usually found in animated sequences, but it will gracefully degrade to standard ray tracing for scenes with pathological animation design.