Optimizations of Cascaded Shadowmapping for Large Terrains

Abstract

Shadowing effects greatly improve the realism and the perception of virtual 3D environments by providing useful visual cues. But due to their global nature, a high computational effort is needed. Shadowmapping is one of the most popular approaches for real-time rendering. A more advanced Shadowmapping technique often used for larger areas is Cascaded Shadowmapping (CSM). Since it is an image based method it has several aliasing and artifact problems as the rendered data sets get bigger and the terrains get larger. To reduce these problems several optimizations are proposed, described and evaluated in this thesis: Most applications set the near and far plane of the view frustum to static values. By calculating these plane values dynamically, the sampling rate of each shadowmap is highly improved as the texture space is used more efficiently. Furthermore, each cascade is extended towards the negative light direction in order to handle missing shadows which originally result from wrapping the cascades around the view frustum. Shadow flickering is another issue which is caused by dynamically calculating cascade sizes. This is highly reduced by moving the cascade sizes in texel steps and changing its size in custom discrete steps. A Cascaded shadowmap specific problem poses overlapping areas. By discarding fragments of these overlapping areas in the larger cascades, it is guaranteed that every object is only represented in a single shadowmap. These optimizations are described and tested by using a Terrain Renderer which is mapping topography to a sphere to visualize planets. This Terrain Renderer is using Level of Detail and dynamic data loading. Thus, these concepts are also taken into account. The results show that the shadow quality and stability is highly increased while not compromising the performance Shadowmapping is providing. It can be used for shadowing large terrains that are illuminated by the sunlight.