Fusing Environment Maps for IBL in Large Scenes
Author
Pedersen, Jesper
Term
4. term
Publication year
2008
Abstract
Denne afhandling adresserer udfordringen med at udvide store, virkelige scener med virtuelle objekter, hvis udseende matcher scenens rumligt varierende belysning. Med udgangspunkt i en gennemgang af environment mapping, billedbaseret belysning, billedbaseret rendering og global illumination foreslås en metode, der opfanger den varierende belysning med flere HDR-lysprober og bagprojekterer dem på en grov geometrisk model (geometrisk proxy) af scenen. Ved kørselstid approksimeres den lokale lysfordeling ved det virtuelle objekts placering ved at gengive et miljøkort baseret på de fusionerede prober, som derefter bruges til at give materialer korrekt belysning og kaste skygger i overensstemmelse med omgivelserne. Arbejdet behandler praktiske forhold som parallakseforvrængning, valg af samplingstrategi og sammensmeltning af målinger fra flere prober. Alternativer til repræsentation af lysfordelingen, herunder rumlige og frekvensbaserede tilgange, sammenlignes. Implementeringen skitserer udnyttelse af GPU og summed area tables for at accelerere beregninger. Resultater og evaluering omtales i den fulde afhandling, men specifikke udfald fremgår ikke af dette uddrag.
This thesis addresses the challenge of augmenting large real-world scenes with virtual objects whose appearance matches the scene’s spatially varying illumination. Building on a review of environment mapping, image-based lighting, image-based rendering, and global illumination, it proposes a method that captures varying illumination using multiple HDR light probes and back-projects them onto a coarse geometric proxy of the scene. At run time, the local radiance at the virtual object’s position is approximated by rendering an environment map derived from the fused probes, which is then used to shade materials and cast shadows consistent with the surroundings. The work discusses practical issues such as parallax distortion, sampling strategy, and the fusion of measurements from multiple probes. Alternatives for representing radiance, including spatial and frequency-domain approaches, are compared. An implementation outline includes GPU utilization and summed area tables to accelerate computations. Results and evaluation are presented in the full thesis, but specific outcomes are not included in this excerpt.
[This summary has been generated with the help of AI directly from the project (PDF)]
Documents