Thesis of Thibault Dupont


Subject:
Representation, modelisation and procedural amplication of natural riverscape

Defense date: 03/12/2019

Advisor: Eric Galin
Coadvisor: Adrien Peytavie

Summary:

The creation of realistic virtual landscapes is a major challenge in computer graphics. The aim of this thesis is to amplify a large terrain and add dynamism by animating pseudoperiodic phenomena. That is why we became interested in scenes involving rivers: modeling this kind of scene includes the generation of the river network geometry, but also the animation of the corresponding water movement. Although these two elements have been studied individually, there is no method to generate the network and the animation concomitantly. In order to solve this problem, we propose a new river model allowing the animation of the water surface as well as a procedural generation method to sculpt and animate a coherent river network on an existing site.
Our river model uses a hierarchical construction tree to form the riverbed geometry and we propose a new model called blend-flow tree to represent the water surface. The leaves of the tree, the flow primitives, are spatially localized and time-varying on compact support. The functions represent specific characteristics such as ripples, turbulences or eddies. They are procedural and configured so they can generate variety. The internal nodes of the tree are combination operators that form the final surface of the water. This model allows to create a real time animation of the water on rivers several kilometers long while including details locally.
We also propose a river generation method that uses the elevation of a bare terrain to obtain the flow network. It then digs the riverbeds in the ground and simultaneously generates the animation of the water surface. Characteristics, such as the width, depth and shape of the riverbed, as well as the elevation and flow of the fluid surface, depend on the terrain and the type of the river. The riverbed is excavated in the ground by combining compact supported elevation primitives sampled along the river's path. The generation allows the production of a wide range of river shapes, from meandering rivers on the plains to mountain rapids. The model allows an interactive and intuitive editing of river trajectories but also of the choice of primitives. Our river creation method allows us to amplify a large area of land by sculpting a precise riverbed but also to create the animation of the corresponding water surface.