Accelerating Ray Tracing by Exploiting Frame-to-Frame Coherence

As 3d data sets become increasingly larger, and image sizes stay relatively constant, new rendering paradigms based upon images begin to offer significant advantages over traditional object based systems. Current image based systems, however, often assume a lack of the true 3d geometry, and so are limited to the views they can reproduce. We present an image space acceleration method for rendering novel views of large data sets by exploiting frame-to-frame coherence. By reusing visibility and rendering information calculated in the pixels of previous images, we can render most pixels in a new image independently of the 3d object data set. We assert that the majority of pixels within images comprising animations or walkthroughs can be rendered in this manner, and show significant speedups on ray traced animations.