Résumé
In the past decade, with the power of graphical boards, visualization of virtual natural scene become popular in multimedia applications. It usually relies on degraded virtual single plant geometrical representations and massive use of textures. Such scenes show however poor variability in terms of the number of species and individual plasticity. They also show strong animation constraints, many limited to viewpoint moves, ignoring growth processes. We propose here a frame for future works related to virtual plant visualization. It aims to drop down the classical geometrical descriptions of individual plants for the benefits of functional representations, aggregated up to a single crown. We then show that a wide range of parametric virtual crown shapes can be built from the knowledge of leaf area. Moreover, production outputs computed at each simulation time step by crop models or FSPM models define strong constraints in the organ positioning. Such constraints can be used to build hierarchical virtual geometries underlying the plant main axis and its crown decomposition according to its axis typology. Conversely, similar representations can also be generated from an exhaustive representation of virtual or real trees. On such plants, we first build a point cloud from the organ 3D coordinates. A statistical hierarchical dynamic clustering analysis can then be applied to the leaf cloud. It allows obtaining the statistical ellipsoid decomposition that can then be used for comparisons or shape fitting optimization. We finally introduce some technical elements showing that basic 3D shapes used for functional visualization (cone frustum and ellipsoids) can be fully generated and rendered by GPU techniques. As a summary, simulated plant functional visualization appears as a promising research track, freeing models from complex and costly geometrical computations. These representations also propose a new frame of discussions on tree crown modeling and descriptions for diagnosis purposes.