Análise comparativa de diferentes técnicas de iluminação volumétrica

Abstract

This work implements, analyzes, and compares three real-time volumetric lighting rendering methods based on distinct approaches: image post-processing (Screen Space), space subdivision (Froxel), and geometric generation (Polygonal). Volumetric lighting is an optical effect that occurs when light interacts with particles in the transmission medium, such as dust or fog, producing visible light beams. A graphics engine was developed in C++ using the Vulkan API, capable of rendering 3D mesh scenes with multiple directional lights and interchangeable volumetric lighting techniques. The application automatically captures screenshots and records metrics such as rendering time and memory usage. The image post-processing method offers low computational cost but suffers from visual limitations like lack of occlusion and sampling artifacts. The Froxel method subdivides the camera frustum into 3D cells (froxels), accumulating light along view rays using 3D textures and GPU parallelism, balancing performance and quality. The Polygonal method uses geometry generated from shadow maps to represent the light volume with high visual detail at a higher computational and memory cost. Results indicate that the Froxel method achieves the best runtime performance and scalability with scene complexity, while the Polygonal method delivers superior visual fidelity when resources allow. The image post-processing method stands out for implementation simplicity and low memory usage, suitable for hardwareconstrained scenarios. All three methods were systematically evaluated across multiple scenes, and the full source code, experimental data, and test scenes are publicly available to ensure reproducibility and facilitate further research.