Realtime Physically Plausible Global Illumination Using Scalable Spherical Harmonics Hierarchies

Jonathan Metzgar
PhD Dissertation
Published 2018

Sudhanshu Semwal (Chair), Tim Chamillard, C. Edward Chow, Anatoliy Glushchenko, Yanyan Zhuang

Realtime Physically Plausible Global Illumination Using Scalable Spherical Harmonics Hierarchies (PDF)


Global illumination that is based on physically based reflectance models using unbiased statistical path tracing methods remains the best way to simulate a realistic image, but only a few types of simple scenes can be used interactively. The alternative is to produce an estimate of the indirect illumination. Several methods such as virtual point lights (VPL), irradiance volumes / precomputed radiance transfer (IR/PRT), and voxel cone tracing (VCT) provide estimate methods, but require either special precomputation or hundreds to thousands of local point lights. In this dissertation, we introduce a new algorithm called the Scalable Spherical Harmonics Hierarchies (SSPHH) technique which produces a physically plausible estimate of both direct and indirect illumination by introducing spherical harmonic encoded local dynamic radiance maps, called spherical harmonic lights (SPHLs) which are sampled from low or high fidelity path traced radiance probes. Its advantages include small numbers of light probes and no additional requirements on geometry specification. Artists place spherical harmonic lights in a 3D environment. The SSPHH algorithm proceeds through four phases of initialization, visibility determination, radiance probe generation, and hierarchy construction. We use adjacency information to simulate light transport between radiance probes based on statistical measurement of visibility. We control the depth of light transport by sorting and limiting the number of adjacent nodes according to highest contribution. Finally, we contrast the SSPHH algorithm with the process of VPL, PRT, and VCT algorithms. Statistical and error analysis based on reference image comparison, and parameter optimization based on time analysis are also presented.