View-adaptive Importance Sampling for VAL-based Global Illumination (DAAV)

The content of this thesis is to extend an existing real-time global illumination technique that uses Virtual Area Lights (VALs) for indirect illumination in combination with point-based shadow maps, so called Imperfect Shadow Maps (ISMs) to generate corresponding soft indirect shadows. A clustering algorithm inspired by k-Means creates VAL-clusters in light space, which are updated on a frame-by-frame basis. In this talk we will present concepts and techniques to eliminate some of the constraints and limitations of the existing solution as well as incorporating new ideas to make the algorithm more robust for larger, more complex and dynamic scenes.