CG Research Projects

Research > Projects > Real-Time Rendering

Global illumination has been used for generating photo-realistic pictures in different application areas. One way to solve for the global illumination is to use Monte Carlo techniques. In this project, we study various aspects in the Monte Carlo global illumination methods. We mainly concentrate on two aspects:

A New Adaptive Density Estimator for Particle-Tracing Radiosity

In particle-tracing radiosity algorhtims, energy-carrying particles are traced through an environment for simulating global illumination. Illumination on a surface is reconstructed from particle "hit points" on the surface, which is a density estimation problem. We proposed a new orthogonal series estimator to tackle the problem. In the new method, the appropriate number of terms that should be used in the series is determined adaptively and automatically. Moreover, a surface subdivision scheme is combined with the estimator to increase the accuracy of estimation. The new method has several advantages over other existing methods: (1) it requires less memory than the adaptive meshing method; (2) it does not store all the particle-hit points as in the kernel method; (3) it determines automatically how many terms should be used in the orthogonal series; (4) it incorporates surface subdivision to further increase the accuracy of estimation.

Parallel rendering

In addition to find new global illumination algorithms, we believe that such new algorithms should also be easily parallelized in order to make use of nowadays parallel processing power. Based on today's technology, a single processor computer does not have the capability to achieve real-time global illumination. Even though the processor speed will become faster in the future, the complexity of the simulated scene will increase as well. On the other hand, more and more powerful multi-processor machines or multi-computer clusters are commercially available. How to make use of this parallel processing power is another major challenge when designing new graphics algorithms.