A Hybrid Technique for Creating Meaningful Abstractions of Dynamic 3D Scenes in Real-Time

Abstract

We present an approach for creating non-photorealistic renderings of 3D scenes in real-time. We employ a hybrid system which uses both image-space and object-space techniques for creating fast and effective results. A reliable method of edge detection is presented to find all important edges within a scene. This edge detection technique is combined with a painterly renderer to render the scene using different levels of abstraction. This is used to increase the saliency of important objects and remove extraneous detail. 3D object information is used to apply an object-based segmentation technique which allows each scene object to be rendered using a single abstraction level depending on its scene importance. The abstraction techniques are implemented on the GPU which helps the system achieve interactive rates.