Efficient Processing of Plant Life in Computer Graphics

Abstract

The role plants play in our daily lives gives synthetically modeled trees and plants particular significance. Vegetation is part of almost all virtual scenes and also in various application domains. The inhomogenous structure of these natural objectsyields an enormous geometric complexity, which continues to pose challenges to computer graphics researchers to this day. This thesis presents new techniques, methods and paradigms that contribute to the field of processing vegetation in computer graphics. Each of the following three approaches helps to migrate the so far mostly static tree models to dynamically adapting and developing ones.A novel tree modeling paradigm is presented, which abstracts a given tree model into a light-weight intermediate representation. This data structure is easy to store and transmit and allows faithful reconstruction of a wide range of different models. Existing tree models as well as laser-scanned point sets can be converted into this representation. This either enables compression of the data necessary to describe a model, or to reconstruct exemplars from point sets. When defining virtual sceneries, artists, designers or architects often need to adjust tree models according to the constraints of a scene. This is a requirement that is not addressed by today's modeling approaches.A novel interaction technique is presented, which allows trees to react dynamically to changing environmental conditions. The main branching structure of a given model is deformed based on transformations applied to the underlying skeletal graph. The new technique maintains a biologically plausible structure and is efficient enough to be used in interactive and real-time applications.Another new method is introduced, which allows the developmental stages of static input tree models to be computed. Besides creating a large variety of exemplars from a single input model, the interpolation of these stages enables the tree's natural growth to be replicated. As an input tree model does not contain all the branches the tree will have developed during its entire lifetime, a new approach is presented, which allows additional branches to be generated for the younger developmental stages. This process utilizes the structural similarity of the input and is controlled by several measures known from botanical research.Finally, techniques are explored, which enable vegetation to be rendered in real-time. This includes approaches for generating the mesh of branching structures as well as means for maintaining a photo-realistic appearance.