Abstract
Rendering global illumination effects for dynamic scenes at interactive frame rates is a computationally challenging task. Much of the computation time needed is spent during visibility queries between individual scene elements, and it is almost illusive to update this information at realtime even for moderately complex scenes. In this paper, we propose a global illumination approach for dynamic scenes that runs at near-real-time frame rates on a single PC. Our method is inspired by the principles of hierarchical radiosity and tackles the visibility problem by implicitly evaluating mutual visibility while constructing a hierarchical link structure between scene elements. By means of the same efficient and easy-to-implement framework, we are able to reproduce a large variety of complex lighting effects for moderately sized scenes, such as interreflections, environment map lighting as well as area light sources.
Introduction
Global Illumination (GI) is essential for achieving photorealism in computer graphics, accounting for light that bounces between surfaces—such as color bleeding from a red wall onto a white floor. However, the primary bottleneck for GI is visibility: determining if light can travel between two points without obstruction. Traditional methods like Radiosity or Ray Tracing are computationally expensive due to millions of visibility queries. While Precomputed Radiance Transfer (PRT) offers real-time performance, it typically requires static geometry. This paper introduces a method for computing full GI for dynamic and deformable scenes at interactive speeds on a single PC without heavy precomputation.
Methodology
The proposed approach builds on the principles of Hierarchical Radiosity but introduces "Implicit Visibility" to bypass expensive ray-traced checks.
- Geometric Hierarchy: The scene is represented as a hierarchy of "surfels" (surface elements). At the finest level, these are based on mesh vertices; coarser levels represent clusters of geometry.
- Discretized Directional Bins: For each surfel, the surrounding sphere is divided into bins using a cube-map structure.
- Implicit Visibility via Link Building: Instead of casting rays, the algorithm builds links between elements. When a link is proposed for a directional bin, the system only retains the link to the closest surface element. This effectively treats visibility like a shadow map for every point, resolving occlusion "for free" during link construction.
- GPU Acceleration: The entire pipeline—hierarchy updates for deforming meshes, link building, and light gathering—is implemented on the GPU to maximize parallel processing.
Results and Evaluation
The method was evaluated using several standard models, including a Teapot (3,878 vertices), a Flying Dragon (2,670 vertices), and a Monster (3,378 vertices).
- Performance: On an NVIDIA 8800 GPU, the system achieved 4–12 FPS for scenes with 3,000 to 5,000 vertices. The algorithm scales at $O(N \log N)$, making it significantly more efficient than traditional $O(N^2)$ radiosity.
- Visual Fidelity: The method successfully reproduces complex effects like color bleeding and soft shadows. While slightly softer than ground-truth path-traced images due to directional discretization, the results are visually comparable.
Discussion and Future Work
Despite its efficiency, the method has limitations. Coarse directional bins can lead to "blocky" artifacts or light leaking, while very coarse meshes may allow light to leak through surfaces. Additionally, while it supports glossy reflections, sharp mirror-like reflections would require a prohibitively high bin resolution. Future work could explore temporal coherence to reuse information from previous frames, further accelerating the rendering process.
Summary updated