Abstract: To avoid needless processing of non-visible portions of a 3D scene, an occlusion culling method which dynamically identifies large occluders as the viewpoint moves is applied to discard non-visible polygons early in the graphics pipeline, i.e., in object space. The method is conservative, i.e., it overestimates the set of visible polygons; it exploits spatial coherence by using a hierarchical data structure; and it exploits temporal coherence by reusing visibility information such as supporting and separating computed for previous viewpoints. Objects are called from a three-dimensional object database if they are fully occluded from sight of the viewpoint due to either occluders near the viewpoint or occluders near the objects themselves. Objects which have not been called are drawn. Supporting planes tangential to an occluder and object set are used to determine occlusion.