Rendering-Based Optimization for a Near-Eye Display and Active 3D ScanningPublic Deposited
Raytracing is a long-established means to simulate physically accurate light propagation. Increasing availability and power of highly-parallel computing, such as cloud-based clusters and dedicated graphics hardware, means that rendering algorithms can produce high resolution output very quickly. This means raytracing can now be used as a forward model in optimization algorithms to improve the performance of computational imaging systems. This thesis considers the hypothesis that this rendering-based optimization approach for computational imaging systems will gain increasingly-widespread use in the future. This thesis evaluates two distinct uses for rendering-based optimization: a near- eye display and a surface reconstruction algorithm for active 3D scanners. In the first, a novel display architecture produces spatially varying focus for the user. Rendering-based optimization corrects hardware-induced optical distortions to produce an improved retinal image. The second use case flexibly corrects for multibounce interference in active 3D scanners, including arbitrary scene reflectance, using rendering-based optimization.