Comparison of two Methods of Global Illumination Analysis

Contents Introduction Overview of Deterministic algorithm Overview of Monte Carlo algorithm Scenes used for comparison Accuracy analysis Results obtained Conclusion Acknowledgments

1. INTRODUCTION

The goal of this paper is experimental comparison of two methods for finding global illumination distribution. The two methods compared are the Deterministic Radiosity method and the Monte Carlo method based on Forward Ray Tracing. The main criterion of the comparison is accuracy vs. processing time. Also perceptual quality is considered as an additional criteria.

Turbo Beam Tracing (TBT) software developed by INTEGRA, Inc. was used to actually do the experiments. Both the methods mentioned are implemented in TBT. Some details of their implementations important for understanding of results are overviewed in ch. Overview of Deterministic algorithm, and ch. Overview of Monte Carlo algorithm. TBT with the Monte Carlo analysis and other extensions to support lighting analysis and design has the commercial name SPECTER.

In general, the TBT model to run lighting simulation consists of:

Geometry: defines shapes of the objects to be rendered. Actually in our experiments only sets of triangles were used.
Attributes: define properties of the materials, of which the objects are built, such as color, light reflecting / refracting properties etc.;
Lighting: defines light sources illuminating the model, their intensities, spatial distribution of illumination etc.;
Viewing parameters (synthetic camera): decides among others, what fragment of the rendered model will appear on the image.

Such bundle of data is called a "scene" in computer graphics.

We used several scenes to compare the Deterministic and Monte Carlo algorithms. These scenes are described in ch. Scenes used for comparison. For each scene we executed several runs of both the methods observing how the accuracy achieved depends on run time. Among the scenes used there is one very simple artificial scene for which we can find reliable and accurate estimates of the results (luminance distribution). Having this reference solution we can provide absolute accuracy analysis.

Other scenes are more or less complex; they describe real interiors. For these scenes there is no precise reference solution, so accuracy was estimated by comparing each simulation result (presented as distribution of luminance in the image plane) with the most accurate one. Our approach to accuracy analysis is described in ch. Accuracy analysis.

Finally, ch. Results obtained presents the results of the comparison (time / accuracy graphs for both methods) and ch. Conclusion discusses advantages and disadvantages of each method revealed by this investigation.

Contents Introduction Overview of Deterministic algorithm Overview of Monte Carlo algorithm Scenes used for comparison Accuracy analysis Results obtained Conclusion Acknowledgments


Copyright 1996 Andrei Khodulev.	[Home]