Stochastic level of detail in computer animation

Abstract

A method for smoothly transitioning between different object representations in computer animation using stochastic sampling. The method allows for level of detail transitions between object representations made up of different geometric primitives, of different types, with different rendering attributes, and even different topologies without “popping” or other visual artifacts.

Description

BACKGROUND OF THE INVENTION[0001]The invention relates generally to the art of computer graphics and computer generated animation. More particularly, the invention relates to the use of, and means for smoothly transitioning between, different representations of objects depending on the object's visibility and importance in the rendered scene.[0002]In computer rendering (digital image synthesis) objects in the synthesized image must be mathematically represented in three dimensional object space. This is achieved by modeling the object's bounding surfaces as a collection of geometric primitives. Typically, the primitives are simple polygons or more complicated surface elements defined by non-linear paramatized curves (e.g. NURBS (Nonuniform Rational B-Splines)).[0003]The realism obtainable in the resulting image depends to a large degree on the number and complexity of the primitives used to represent the objects. The flip side is that more., and more complex primitives require more ...

AI Technical Summary

Benefits of technology

[0010]By combining the techniques of the present invention with those of Cook et al. one can, in a unified way without post processing or pixel level manipulations, produce smooth efficient animation incorporating multiple levels of detail, antialiasing, motion blur, depth of field and soft shadows in which visibility is correctly determined at each sub-pixel sample location.

[0011]Because in the present method individual samples are evaluated against a single LOD representation for each object, and visibility is computed correctly for each subpixel sample, it is more efficient and produces better images than can be obtained by cross-dissolving at the pixel level images separately rendered from different object representations. In addition, because the present invention does not depend on, or constrain, the details of the geometric representations of the objects, it allows one complete freedom in the definition and representation of objects. One can, for instance, transition between entirely different representations of individual objects, e.g., a highly detailed, texture mapped and trimmed NURB representation of a leaf on the one hand, and a green square on the other.

[0012]Moreover, one has complete freedom in defining the “object” for LOD purposes, and one is free to vary the definition throughout the animation. For example, one may create and store a hierarchy of different LOD representations of a forest, a tree in the forest, a branch on the tree, or a leaf on the branch, and choose independently and consistently importance criteria and detail ranges for “objects” in each level of the hierarchy. In one group of scenes the rendered object may be a tree represented by one or several LOD representations. In another group of scenes the rendered object may be a leaf on the tree. In both cases the present invention allows one to incorporate and smoothly transition between different LOD representations of either object.

Problems solved by technology

In both the flight simulators and Funkhouser and Sequin walkhrough, the transition between object representations is instantaneous and discrete resulting in “popping”, a visual artifact that is unacceptable for high quality computer animation.

None of the prior methods allows one to create smooth transitions between representations with arbitrary modeling primitives, topologies, and shading paradigms, including smooth transitions between arbitrary three dimensional geometric representations and approximations of them using displacement or texture maps.

This technique is inefficient, requiring multiple renderings of each object, and results in poorer image quality because the visibility computation is only approximate at the whole pixel level and does not fully account for antialiasing, reconstruction filters, or motion blur already applied to these pixels.

Images