Global illumination, e.g. the determination of the amount of light received by objects in a virtual scene, is an important step in the realization of realistic computer images. It provides indeed, through the use of shadows, color bleeding, indirect light, the clues that make an image look real. Hierarchical Radiosity algorithms are usually cited as efficient methods to solve this problem because of the inherent multi-level approximation of the light distribution they use. However, in really large scenes, which may include millions of geometric primitives, Hierarchical Radiosity Algorithms become limited by the memory cost of storing thoses primitives. Indeed, the complete list of objects in a scene seems to be required at the same time in memory to be able to participate to the lighting of any other part of the scene. Using an alternative representation of objects and by sharing their lighting properties at a high level in the scene hierarchy, this piece of work has shown that it still is possible to apply such a global illumination algorithm while only keeping a small part of the scene in memory at a time. Practical examples of applications proove the efficiency of the method in various contexts such as interior scenes or complex vegetation lighting.

Geometric models of vegetation have been produced by a various software packages for a number of years. These models usually represent realistic looking trees and plants at a user-defined age of growth. While this kind of models satisfy the computer graphist for a certain class of problems, the fact that the generated plant is only a static set of geometric primitives is a limitation for some applications: the models do not fit to the geometry of the scene they are included in and the plants do not have the shape they should have due to the presence of nearby light sources. We have proposed a new model of plant generation that reproduces its internal functioning and computes its growth as a side effect. The simulation accounts leaf transpiration and photosynthesis activity to compute the amount of vegetal matter fabricated at any time during the growth and allocates this matter according to a clever process among the different organs. While the growing model decides of the specy of plant we get, the lighting conditions influence the resulting shape of the plant. We have developped a proper lighting simulation algorithm to cope with the specific difficulty of vegetation scenes, using which we are able to grow plants that react to their lighting environment just like real plants would do. Besides, the plant growth model being callibrated with real agronomic experiments, it can be used to compute and optimize agronomic parameters without wasting the time and money needed by an on-site experiment.