Simulation of Clothes
for Real-time Applications

N. Magnenat-Thalmann, F. Cordier, M. Keckeisen, S. Kimmerle, R. Klein, J. Meseth

Keywords

Real-time graphics, real-time cloth deformation, collision detection, virtual heritage, numerical solvers, real-time rendering of textiles, data-driven simulator.

Overview

Simulating clothed virtual humans in real-time is required in many applications such as enhanced reality, video games, virtual heritage... This course gives a description of the latest state of the art for real-time cloth simulation. The course will present the following concepts:

  • The creation of garments.
  • Techniques for collision detection
  • Numerical solvers
  • Simulation of the cloth behaviour
  • Real-time rendering of textiles

Several case studies will be shown: the simulation of the Roman life in Pompei, reconstruction of Aya Sofia church in Turkey, the Virtual Try-On...

Introduction and State of the Art in Clothing

This presentation gives an overview of the whole tutorial. It introduces the different techniques that will be presented in the next sessions. An overview of previous work is given as well.

  • Overview of the tutorial
  • Problematic of simulating clothes in real-time
  • Motivating applications
  • Overview of the real-time cloth simulation techniques:
    • Physics-based simulation
    • Collision detection
    • Example-based approaches

Physical Models & Numerical Solvers

Physical models and numerical solvers are the basis of current cloth simulation engines. We will discuss physical models for cloth ranging from simple mass-spring systems to continuum-based particle systems and fast finite element solutions. Then, we will describe several explicit and implicit time integration schemes and compare them with respect to stability and performance.
  • Physical models
    • Discrete models
    • Continuous models
  • Numerical simulation
    • Methods for numerical integration
    • Solving nonlinear systems
    • Comparison of methods Break

Collision Detection

In the simulation of cloth, collision detection is crucial for the quality of the results and for the performance as the collision detection process can be highly complex. We will first discuss the specificity of collision detection for clothes and will thereafter review some recently used techniques for the real-time collision detection of clothes. In the last section we will detail a fast collision detection approach based on bounding volume hierarchies as it is specially suited both for animated as for static scenes and is also able to detect self-collisions.

  • Techniques for collision detection on deformable models
    • Distance fields
    • Image-space techniques
  • Bounding volumes hierarchies
    • Hierarchies and spatial subdivision
    • Hierarchy traversal and self-collision heuristics
  • Detection strategies for deformable objects
  • General collision detection on polygonal meshes
  • Self-collision detection
  • Data-driven collision detection

Real-time Rendering of Textiles

Reproducing the complex reflectance properties of complex materials like fabrics at realtime frame rates is an important part of cloth visualization systems. In this part, we will present methods for acquisition, compression and real-time rendering of material reflectance represented as bidirectional texture functions (BTFs). The presented rendering approaches include large-scale shadows and image-based lighting leading to a highly realistic appearance of visualized clothes.

  • BTF Acquisition and Compression
    • Automatic HDR BTF acquisition
    • Compression techniques
  • High-quality and real-time BTF rendering
    • Hardware-accelerated
    • Image-based lighting
    • BTF ray-tracer

Context-Specific Cloth Simulation

Most of the existing approaches use a general-purpose simulation method using collision detection and physical simulation for the whole garment. Unfortunately, simulations that simply calculate all potentially colliding vertices may generate a highly realistic movement, but do not provide a guaranteed frame time. This talk will demonstrate how the computation cost can be greatly reduced by making use of predetermined conditions between the cloth and the body model, avoiding complex collision detection and physical deformations wherever possible.

  • Simulation of the cloth dynamics with geometric and physics-based models
  • Geometric deformation of cloth wrinkles
  • Data-driven clothes simulation

Case-studies in National and European Research Projects

This session will show several applications of cloth simulation in the frame of European and National Research projects:

  • Life Plus
  • Cahrisma
  • E-Taylor
  • RealReflect
  • Virtual Try-On

Presenters

Nadia Magnenat-Thalmann (thalmann@miralab.unige.ch) has contributed to pioneer research into Virtual Humans for over 25 years, and has participated to spectacular stateof- the-art demonstrations and rigorous and intensive academic research programs that make them possible. After her PhD in Quantum Physics and Computer Graphics from the University of Geneva, she has been a Professor at the University of Montreal in Canada from l977 to 1988. In l989, she founded MIRALab, an interdisciplinary Research Lab at the University of Geneva in Switzerland.

Frederic Cordier (cordier@miralab.unige.ch) received his PhD in Computer Science from MIRALab, University of Geneva in January 2004. His current research interests include real-time animation of virtual humans, particularly on real-time clothes, skin deformation and interaction among clothes and skin. In this area, he has published several journal papers in Computer Graphics Forum, IEEE Computer Graphics & Applications, Journal on Visualization and Computer Animation, etc.

Pascal Volino (pascal@miralab.unige.ch) is a computer scientist, working at MIRAlab, University of Geneva. He is actually working on new models for cloth animation, involving versatile models for efficient simulations on situations involving high deformation, wrinkling and multilayer garments. The research is particularly focused on data structure, efficient collision detection, robust simulation and interactive cloth manipulation.

Michael Keckeisen (keckeise@gris.uni-tuebingen.de) studied Mathematics and Computer Science at the University of Tübingen and received a Diploma in Mathematics (2000). Since 2001 he is a member of the graphics research group at the University of Tübingen. His main research interests are the physically based modelling and simulation of clothes, and interaction techniques in virtual environments.

Stefan Kimmerle (kimmerle@gris.uni-tuebingen.de) studied physics and chemistry in Tübingen and San Diego. In 2000 he received his Diploma in physics from the University of Tübingen. Since 2001 he is a Ph.D. student at the graphics research group at GRIS. In 2003 he was an invited researcher at GRAVIR, INRIA Rhône-Alpes in Grenoble. His main research interests are physically-based modelling and collision detection for deformable objects.

Reinhard Klein (rk@cs.uni-bonn.de) received his Ph.D. in computer science from the University of Tübingen, Germany, in 1995. After receiving an appointment as lecturer ("Habilitation") in computer science in 1999, he became an Associate Professor at the University of Darmstadt, Germany. Since October 2000, Reinhard Klein is full professor at the University of Bonn and head of the department of Computer Science II.

Jan Meseth (meseth@cs.uni-bonn.de) is a research assistant and Ph.D. student in the Computer Graphics Group at the University of Bonn, Germany. His main research interests are photo-realistic, image-based rendering and level-of detail representations.