Introduction

Building realistic models of plants is not a simple task. Such models generally have the following caracteristics:

- A large number of small polygons. A single palm tree of the the image shown above, for instance, requires between several thousands of polygons.

- A structure that obeys some specific rules, that determines the particular species the plant belongs to.

- A notion of multi-scale self-similarity.

Considering these requirements, three methods seem adapted for obtaining models of plants: scanning real plants, L-systems and simulation of plant growth.

- Scanning, e.g direct data acquisition, through 3D scanners, can not be considered, because of the topological and geometrical complexity of plants. However, real plants can be used to build models of the radiometric properties measured of the leaves.

- L-systems are fractal-based plants emerging from simple sets of mathematical rules. This technique allows to build nice models of plants, but does not allow to account for the influence of natural factors on plant growth, in a biological manner.

- biological simulation of plant growth, as provided by the AMAP software is the most promising technics in terms of realism, because it allows to account in the simulation for every physiological factor like minerals and water supply, and the amount of light received by the plant.

Goal

   The aim of the SOLEIL project is to simulate the action of light energy on plant development. This is be done through the collaboration of a radiative exchange simulation program, called BRIGHT, and the plant growth simulator AMAP. Up to know, AMAP is based on the asumption that the amount of light energy received by each leaf is uniform on the entire plant. The new system uses the BRIGHT software to determine the exact amount of light received by each leaf in the plant, thus considering the orientation of the leaves and macro effects like self-shadowing. AMAP incorporates this lighting information to determine the production inside each leaf, thus modifying the geometry of the model. The figure below shows the process:

  Applying the radiosity method to the simulation of plant growth needs to solve some specific problems, like

- accounting for leaf partial transparency

- accounting for transparency and reflectance models for different wavelengths: infra-red, visible light, UV.

- dealing with very large geometric data bases. A single plant can be made out of thousands of polygons. The use of multi-scale representations of the geometry (clusters) greatly helps solving this problem.

This new functionnality opens the way to studying different phenomena, at various scales:

• At each leaf, the amount of light received allows to compute both the temperature of the leaf (hence the leaf perspiration and the speed of the different chemical reactions into it) and the amount of light energy received, which directly rules the photosynthetic reactions that create the body of the plant.

• At a higher level, some macro phenomenas rule the shape of the entire lant:
• phototropism (deformation of the plant to maximize the light energy received
• flowering
• deformation of some part of the plant due to local lack of energy (self-shadowing)
  
• At the level of an entire plantation, some competition phenomena appear between the different plants.
  
  
  

Related problems

• Efficiently computing radiative exchanges between large sets of polygons requires the use of clustering methods: the scene is expressed as a hierarchy of bounding volumes starting from each polygon, up to the entire scene. Constructing intelligent hierarchies of clusters for deeply interlocked models of plants is not simple.
• Light sources are supposed to move constantly, as the sun for instance, changes its position during the day. This asks for instance the question of using the coherence of the various positions of the sun to efficiently compute a solution.
• In image synthesis, the notion of image quality is closely related to human perception capabilities, which makes discontinuities and various artefacts in the images very important. In the context of a botanical simulation, controlling the error takes a different form: as we only need an estimation of the mean value of energy received by the plant, estimating the error to drive the refinment of the computation must be based on specific criteria.
• To cope with the huge amount of polygons, we have designed a special hierarchical radiosity algorithm based on multi-level instanciation. The approximate self-similarity in plant models allows indeed to replace some parts of the geometry by copies of an original one that remains in memory. To the difference of the geometry the light distribution on the instances can not be shared. We solve the problem using a recursive instanciation algorithm that only need to compute a hierarchical radiosity solution between a small number of objects at any time.