Section: New Results

Animating nature

Participants : Romain Arcila, Guillaume Bousquet, Eric Bruneton, Marie-Paule Cani, Julien Diener, Estelle Duveau, François Faure, Benjamin Gilles, Everton Hermann, Franck Hétroy, Olivier Palombi, Cécile Picard, Lionel Reveret, Maxime Tournier, Xiaomao Wu.

Robust finite elements for deformable solids

Participants : François Faure, Guillaume Bousquet.

We have presented at SIGGRAPH 2009 a new approach for the embedding of linear elastic deformable models [10] . Our technique results in significant improvements in the efficient physically based simulation of highly detailed objects. First, our embedding takes into account topological details, that is, disconnected parts that fall into the same coarse element are simulated independently, as illustrated in Figure 10 . Second, we account for the varying material properties by computing stiffness and interpolation functions for coarse elements which accurately approximate the behaviour of the embedded material. Finally, we also take into account empty space in the coarse embeddings, which provides a better simulation of the boundary. The result is a straightforward approach to simulating complex deformable models with the ease and speed associated with a coarse regular embedding, and with a quality of detail that would only be possible at much finer resolution.

Figure 10. A model of a liver with attached vascular system sim- ulated with coarse resolution hexahedra. Our technique models the behaviour of the soft liver tissue, stiffer veins, and much stiffer tumors by taking into account a distribution of materials and the presence of empty regions in the embedding. The complex topologi- cal branching of the vascular system is preserved by superimposing elements.

We have also collaborated with the University of Geneva on finite elements which are both numerically efficient and physically accurate, using and extended St-Venant-Kirchhof model [14] . We have applied it to simulate the anisotropic, non-linear behavior of cloth, as illustrated in Figure 11 .

Figure 11. Virtual prototyping applications require an accurate representation of cloth material behavior for evaluating precisely the stretch forces (color scale) on the garment along particular postures of the character (top). An efficient simulation model is also required for the dynamic simulation of high-resolution complex garments involving several layers of cloth.

High-Performance Simulation of Complex Models

Participants : François Faure, Everton Hermann.

Everton Hermann is a Brasilian Ph.D. student funded by a Cordi grant and co-tutored by François Faure in EVASION and Bruno Raffin in MOAIS. We have proposed a parallelization of interactive physical simulations [20] . Our approach relies on a task parallelism where the code is instrumented to mark tasks and shared data between tasks, as well as parallel loops even if they have dynamics conditions. Prior to running a simulation step, we extract a task dependency graph that is partitioned to define the task distribution between processors. This approach has a low impact on physics algorithms as parallelism is mainly extracted from the coordination code. It makes it non parallel programmer friendly, using domain decomposition or task-based parallelism, as illustrated in Figure 12 .

Figure 12. Two ways of parallelizing a complex scenes: domain decomposition (left) and task-based parallelism (right).

Sound synthesis

Participants : François Faure, Cécile Picard.

Cécile Picard, a PhD. student, was previously in Sophia-Antipolis tutored by Nicolas Tsingos and George Drettakis. She has been with us from July 2008 to July 2009, then moved back to Sophia-Antipolis to complete the writing of her Ph.D. dissertation with George Drettakis. We published a method for sound synthesis in game engines [24] , where we bridge the gap between direct playback of audio recordings and physically-based synthesis by retargetting audio grains extracted from the recordings according to the output of a physics engine, as illustrated in Figure 13 .

Figure 13. Overview of our approach combining off-line analysis of recorded sounds with interactive retargetting to motion.

We have also extended our grid-based approach for robust finite elements to physically based sound synthesis [23] , as illustrated in Figure 14 . The technique performs automatic voxeliza- tion of a surface model and automatic tuning of the parameters of hexahedral finite elements, based on the distribution of material in each cell. The voxelization is performed using a sparse regular grid embedding of the object, which easily permits the construc- tion of plausible lower resolution approximations of the modal model.

Figure 14. An example of a complex geometry that can be handled with our method. The thin blade causes problems with traditional tetrahedralization methods.

Real-time animation of liquids and river surfaces

Participant : Eric Bruneton.

Figure 15. Our river animation and reconstruction model handles the real-time dynamic exploration from landscape-scale to close-scale. Advected particles obey a Poisson-disk distribution in screen space.

Last year, Qizhi Yu (a Marie Curie PhD student supervised by Fabrice Neyret and Eric Bruneton) developped a macroscopic model of rivers, allowing for the real-time visual simulation of a flow from close to far view on very large terrains (see Proland Project). A real-time editable vector description of river boundaries and obstacles is used to define a semi-analytic distance field which is use to derivate a zero-derivative flow. A screen-space Poisson-disk distribution of river particles carrying wave textures is animated and continuously readapted, so as to be space-time continuous (see Figure 15 ). This work has been published this year at the Eurographics conference [17] .

Figure 16. Our Lagrangian texture advection model allows local patches to be deformed and regenerated asymchroneously. This yields a better conformance to the appearant flow and to the texture spectrum properties at the same time.

A research report has also been published this year [33] , on a Lagrangian texture advection model developped last year by Qizhi Yu. Our particles are distributed according to animated Poisson-disk, and carry a local grid mesh which is distorted by advection and regenerated when a distorsion metrics is passed. This Lagrangian approach solves the problem of local-adaptive regeneration rate, provide a better spectrum and better motion illusion, and avoid the burden of blending several layers (see Figure 16 ).

Motion capture and animation of vertebrates

Participants : Estelle Duveau, Olivier Palombi, Lionel Reveret, Xiaomao Wu, Benjamin Gilles.

The ANR project Kameleon has driven several research topics towards the achievement of motion capture and animation of small vertebrates. Based on data collected at the synchrotron, Benjamin Gilles has developed a new method to quickly registered new anatomical model of animals on CT and MRI scan to speed-up the segmentation of the articulated bones. A paper will be published at Computer Graphics Forum journal. Estelle Duveau is continuing her PhD on motion capture from 3D surface flow. This PhD is co-advised by Lionel Reveret and U. Descartes, Paris 5. A new method based on physically-based motion capture has been developed. A clinical study is currently under investigation using techniques developed during the project. The CNES has selected this project to bring the study into micro gravity condition using their zeroG airplane facility. A new set-up adapted to the airplane is starting to be built. The methodology has opened new projects : one is focusing on the modeling of walking bird and its potential to be morphed towards anatomy of bipeds dinosaur to simulate their locomotion. Another one is focusing on the parameterization of quadrupeds locomotion (see below).

Figure 17. 3D surface tracking.

Figure 18. Motion capture and animation - walking condition.

Motion capture of animals in outdoor conditions

Participant : Lionel Reveret.

Projects have been started to develop method adapted to the motion capture of animals in outdoor conditions. A pioneer study has been done for greyhound dogs. The goal is to set-up a scientific campaign to perform motion capture in a wild life reserve in Africa. Another project has been launched to capture motion of penguins in Antartica. Video data for marine turtles have been collected in Barbados, in collaboration with McGill university.

Figure 19. Motion capture and animation of a dog.

Character animation

Participants : Marie-Paule Cani, François Faure, Franck Hétroy, Lionel Reveret.

We have presented a general method to intuitively create a wide range of locomotion controllers for 3D legged characters [8] . The key of our approach is the assumption that efficient locomotion can exploit the natural vibration modes of the body, where these modes are related to morphological parameters such as the shape, size, mass, and joint stiffness. The vibration modes are computed for a mechanical model of any 3D character with rigid bones, elastic joints, and additional constraints as desired, as illustrated in Figure 20 . A small number of vibration modes can be selected with respect to their relevance to locomotion patterns and combined into a compact controller driven by very few parameters. We show that these controllers can be used in dynamic simulations of simple creatures, and for kinematic animations of more complex creatures of a variety of shapes and sizes.

Figure 20. The first four non rigid vibration modes of a dog model with 80 degrees of freedom. These modes can be de- scribed as bounding, back twisting, stretching, and alternat- ing legs, respectively.

A state of the art has been published in the Computer Graphics Forum journal [12] about quadruped animation. It follows a publication as State of the Art Report, presented at Eurographics 2008. A research program is starting with the MNHN to derive a 3D animation controller of quadrupeds from theoretical results obtained at the Museum on the description of quadruped gaits.

Motion capture of trees under wind

Participants : Julien Diener, Lionel Reveret.

Figure 21. Motion capture of trees under wind.

These works are carried on in the context of the ANR project Chene-Roseau. The goal is to validate measurement tool from video to evaluate the risk of breaking of fruit trees under strong wind. Several experiments have been done in collaboration with INRA at Clermont-Ferrand (UMR PIAF). In parallel, a work on modal analysis of tree structure and its application to real-time animation had been developed and has been published at EG09 [6] .

Figure 22. Real-time animation of a thousands of trees using modal analysis.

Modeling motion capture data of human

Participants : Lionel Reveret, Maxime Tournier, Xiaomao Wu, Franck Hétroy.

Figure 23. Modeling motion capture data of human.

Works on mathematical modeling of quaternionic signals arising from motion capture have been investigate within the context of the ARC project Fantastik. These works have lead to two INRIA Research Report and one paper has been published at EG09 [13] . This paper has been awarded as one of the best three papers of the conference. Maxime Tournier has spent 6 months at the U. of McGill working with Paul Kry to integrate physical simulation into the statistical approach. A sparse statistical analysis of motion data has been investigated by Xiaomao Wu in collaboration with Maxime Tournier and Lionel Reveret. A paper has been published at the IEEE Computer Graphics and Applications journal [16] . Finally, works have been done on expressive facial animation with the Psychology Department of the University of Geneva. A journal paper is currently under preparation.

Figure 24. Motion compression using Principal Geodesic Analysis (PGA).

A project with Lionel Reveret and Franck Hétroy has been started on learning and modeling climbing gestures. A first data collection has been done with a 2nd year Ensimag student, Simon Courtemanche, involved in this project both as a computer scientist and a competitor in rock climbing. Simon Courtemanche is continuing this work for his M2R in Computer Graphics. This work is extended as a collaboration with Edmond Boyer from the PERCEPTION team.

Figure 25. Modeling rock climbing gestures.

Processing animated meshes

Participants : Romain Arcila, Franck Hétroy.

Mesh animations, or sequences of meshes, represent a huge amount of data, especially when acquired from scans or videos. In collaboration with the university of Lyon (LIRIS lab), we address the problem of partitioning these sequences, in order to later be able to compress them. We started this year by developing a formalism to describe possible mesh sequences and mesh sequence segmentations [26] . We then proposed a first motion-based segmentation algorithm, which clusters mesh vertices into static, rigid or homogeneously stretched components (see figure 26 ). This technique will be presented at the WSCG 2010 conference.

Figure 26. Segmentation of a horse mesh sequence.