Investigating horizon mapping for real-time soft-shadows on planetary datasets

Kurzfassung

Common shadow rendering approaches often produce unsatisfactory results when used for visualizing large scale planetary surface models. All occluding geometry has to be present in memory during rendering and various possible view-lighting-configurations make the creation of an one size fits all parameterization challenging. Horizon mapping is a technique aiming to solve these problems by precomputing the visible horizon of each surface point, which then can be used to efficiently determine the visibility of the light source and thus the shading status of each point. Here, the horizon is sampled in equally spread azimuthal directions and interpolated in between them. In order to produce accurate shadows for all lighting configurations, the azimuthal directions have to be densely sampled, resulting in large additional amounts of data to be stored. In this thesis we discuss different ways to compress these horizon maps into a more compact representation, that preserves details of the horizon. Specifically, we look into using piecewise linear functions with irregularly placed support points and truncated Fourier series. Additionally, multiple ways to generate soft shadows using data from the compressed horizon maps are explored. The various methods were implemented and evaluated in a standalone application. All presented methods achieve real-time framerates above 60Hz even on low end notebook graphics cards. For generating soft shadows, we achieved good results by approximating the solar disk as a series of rectangles and intersecting these with the reconstructed horizon function. Of the discussed compression methods, truncated Fourier series produce static shadow images of similar quality to the reference implementation, while exhibiting more natural dynamic behavior for changing lighting conditions. Here, we also expect the possibility of further improvement by applying run length encoding to the stored Fourier coefficients.