Improving Ambient Occlusion Using LOD Simplification
Comp 236 :: Course Project :: Spring 2006

By: Jason Yanchuleff

Description (revised):

The aim of this project is first to implement ambient occlusion in the real-time raytracer provided by Christian Lauterbach. The next step is use to the LOD
system that exists within the raytracer to improve the performance of the ambient occlusion.

Goals:

The primary goal of this project is to determine if LOD can be used to increase the performance of ambient occlusion in a real-time raytracer. I will run a
series of benchmarks to determine if my implementation using LOD has better performance than the standard implementation.

Prior Work:

I am currently working with Christian Lauterbach in the computer science department at UNC on developing this project. He has written a real-time
raytracer. I will be working to extend this raytracer to include ambient occlusion and then will attempt to increase the performance of the ambient occlusion
using the LOD system already present within the raytracer.

References:

"Assorted Notes About Ambient Occlusion", Greg Coombe
http://www.cs.unc.edu/~coombe/research/ao/

ReduceM: Fast LOD-Based Ray Tracing of Massive Models, 2006

Update 1 (03/15/06):

I have been discussing the premise of this project with Christian to determine the exact requirements and goals. I have also done research into the topic of ambient
occlusion as well into the real-time raytracer written by Christian. I am currently working on trying to become familiar with the code. The next step which should be
fairly straightforward is to implement basic ambient occlusion.

Final Update (05/09/06): Powerpoint Presentation Link

Initial Implementation:

- Ambient occlusion implemented in real-time raytracer (written by Christian Lauterbach) using path-tracing.
- Sends out one ray to find initial hitpoint.
- Then samples a given number of rays in random directions within the hemisphere from that point and tests for an intersection point.
- Calculates a ratio of available light at the point based on the number of rays which intersect.
- Weights the diffuse light contribution at the point based on this ratio.

LOD Implementation:

- Uses the distance from the camera to test initial rays against higher levels of the kd-Tree to improve performance.
- Also uses LOD on secondary rays.
- Ambient occlusion accuracy can also be adjusted by increasing the distance from the camera to the model for secondary rays.

Results:

 

Conclusion:

- Ambient occlusion is still not real-time, but its close with optimization.
- LOD makes some improvement on rendering times.
- Using ambient occlusion accuracy setting makes significant improvement to rendering times.
- Future research could be done to try using the accuracy method with LOD on other raytracing features such as reflections, refractions, etc.

 

Jason Yanchuleff - 2006
Comp 236