Abstract

Despite the robustness, elegance, and superior asymptotic performance advantage offered by hierarchical radiosity solutions, it seems that it is not frequently used in most basic radiosity systems because of its apparent complexity in both conception and implementation. In this project, I present a clean, simple, detailed, and intuitive explanation and implementation of a simple hierarchical radiosity system upon which further research can be easily conducted.

The actual written paper describing this project can be obtained here pkzipped in MS Word 7.0 format.

• implementation of the hierarchical radiosity algorithm as described by Pat Hanrahan in his '91 SIGGRAPH paper.
• found several places where the algorithm could be restructured for efficiency
• used a simpler and much faster form factor calculation and showed that it is sufficient under the hierarchical scheme. This supports the theory I had that the hierarchical method drives the subdivision in order to increase the accuracy of even crude form factor approximations.
• demonstrated that using a very crude one ray sampling for visibility and a simple differential-to-finite area form factor calculation can be sufficient in obtaining fast and accurate results.
• created a viewing system that demonstrates all aspects of the algorithm at work: gathering and shooting radiosity at various levels, the meshing, the link structure at various levels, etc.
• devised a strategy using OpenGL and GLUT that will facilitate background solver iterations that could place more emphasis on the solution building depending on the level of user interaction. In other words, when the user is interacting with the model, the radiosity solver demands are at a minimum; when the system is idle, the solver continues iterating.
• integrated a very complete viewing system that allows for various levels of user-manipulation of the model (trackball, drive, fly, etc).
• tested and gathered statistics on a wide variety of simple models.
• wrote an elaborate paper fully describing the motivation for the hierarchical method along with a complete description of our implementation.

• sample 1
• sample 2
• sample 3

Throughout this project, I did not place much emphasis on the reconstruction phase of the radiosity solution. Instead, I simply wanted to demonstrate a simple hierarchical radiosity scheme that used only flat shading. Here are several attempts at improving the quality first by just setting lower form factor error tolerances, then by obtaining vertex intensities, and finally by sharing vertices. All of these images demonstrate poor methods of reconstruction and exhibit many artifacts. Nevertheless, they do demonstrate the reasonable accuracy of the algorithm and that if given an accurate reconstruction, we would be able to obtain very high quality output.

Higher quality output and reconstruction sample images

• hierarchical radiosity solver modules: hrsolver.hpp, hrsolver.cpp
• link hierarchy building modules: hrfflink.hpp, hrfflink.cpp
• triangular patch quadtree modules: patch.hpp, patch.cpp
• patch linknode modules: linknode.hpp, linknode.cpp
• vertex node modules: vertnode.hpp, vertnode.cpp
• the OpenGL/GLUT modules: opengl.hpp, opengl.cpp
• camera modules: camera.hpp
• a 3D vector class: vec3f.hpp
• the main driver: main.cpp
• UNIX makefile: makefile
• test models used for demo program: test0.dat, test1.dat, test2.dat, test3.dat, test4.dat, test5.dat, test6.dat, test7.dat

• complete pkzipped demo source code.

Simply download the zip file, decompress in an empty directory, and run hr.exe (requires OpenGL .dlls):
• hr.zip