Frustum slicing

In this paper visibility culling is integrated tightly with an octree data structure. This is done by slicing the frustum in such a way that the minimal distance from the eye to objects in a given frustum slice is twice the minimal eye to object distance of the previous slice. Then, if one has a fixed minimal detail size, i.e. a minimal spatial angle so that objects of lesser angular extensions are not rendered to the screen then, going from one slice to the next, objects must be twice as big in order to be rendered. This corresponds to traversing the octree one level less deeply. Thus the minimal detail size hugely cuts the number of nodes that the rendering algorithm must visit, a fact that becomes even more pronounced when noting that small objects are far more prevalent than are big objects. By splitting the frustum into focal and peripheral frusta and, consequently, splitting frustum slices into focal and peripheral ones, one can further take advantage of detail elision by rendering objects far from the line of sight only to some larger minimal detail size.