Real time progressive radiosity in dynamic environment with multiresolution

Abstract

In this project, a new method is presented to shorten calculation time of progressive radiosity [Cohen88] in dynamic environments. The dynamic object has been simplified into the one with lower resolution by implementing edge contraction algorithm [Hoppe96] [Garland97]. One radiosity cycle contains two significant steps: one is calculating radiosity distribution in the scene and the other one is rendering the scene. The dynamic object in the lower resolution form is used to calculate radiosity and the one in the higher resolution form is used to render the scene. Before rendering, the radiosity in the simple model is transferred to the complex one through vertex hierarchy. The algorithm also maintains the shadow of dynamic object very well by implementing Shadow Form Factor Algorithm (SFFL) [Schoffel95]. The essential aim of this project is to achieve real time progressive radiosity. There are two difficulties to achieve this aim: ?? When the number of patches in the scene increases significantly, the calculation time will become overwhelming. ?? In order to deal with dynamic environments, the radiosities distributed in each cycle must be redistributed. So if the number of iterations increases, the calculation time will become overwhelming as well. In order to solve the first problem, Edge Contraction algorithm is implemented to decrease the number of patches in the scene. However, new criterion of vertex pair selection is defined to suit this project. A new algorithm is also proposed to transfer radiosity from the model with lower resolution to the one with higher resolution. As described above, SFFL algorithm is implemented in this project to deal with shadows. In each cycle, SFFL will add a certain energy to old shadow and remove a certain energy from new shadow and update energy of patches in dynamic object. When the number of iterations increases, the calculation time of SFFL becomes siginificantly unacceptable. In order to solve this problem, improvement is made to this algorithm. Instead of recalculating radiosity distributions in each cycle, the unshot radiosity in iterations with the same shooting patch are summed up and redistributed once to the environment. In this way, when the number of iterations increases, the calculation time will not increase obviously. Besides dealing with shadows, calculating form factor in each cycle also contributes a lot of calculation time. Improvement is also made to progressive radiosity. In some cycle, if the shooting patch is the first time to be an emitter, store the form factor values. In some other cycle, if the patch becomes an emitter again, use the form factor stored in the previous cycle directly to calculate received energy from the shooting patch. In this way, form factor is only calculated once. In this report, a detailed survey of previous works on radiosity, multi-resolution algorithms and algorithms integrating both is included. There is also a detailed discussion on my proposed algorithm. Finally, results and future work are also discussed.