Génération intelligente de modèles procéduraux par prospection des paramètres du L-système

Abstract

Procedural modeling is the alternative by excellence to the excessive use of resources both in terms of graphics designer and in hardware costs. The discoveries in the last two decades meet some of the requirements of complex scene modeling. Nevertheless, the lack of flexibility and the appearance of recurring patterns in a scene are still a concern. In this context, this project suggests a set of new procedural tools which will allow users to edit 3D trees by means of interactive adjustment. Our approach relies on a study with participants to perform automatic adjustment of L-system parameters in a way that the resulting trees are meaningful to the users. Seven L-system grammars were used for the study. We derived geometric and topological metrics, which we inspected during the variations of the grammar parameters. In order to identify the most interesting trees, the differences between the initial tree and the generated trees are evaluated through principal component analysis. The tests have established a correlation between the relevance of the generated trees and their location in the charateristic features space of the difference analysis. Moreover, qualifiers assist the users on the type of available modifications on the initial tree. In addition to the etablishment of a metrics system to mathematically evaluate and compare trees, this project contributes to the procedural editing by providing smart selection based on the derivation sequence. Users can edit tree parts by making a selection which is internally refined based on the derivation tree. The necessary number of regeneration on each subset of the tree is inferred from this derivation-based selection method. Also, an overview of the available edits, related to each qualifier, is available in the proposed interface. The results show that our approach can identify rich varieties of trees based on the measurement of the tree differences.

Date	20 Feb 2017
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Eric Paquette (Supervisor)