Measurements and Models for Motion Capture

During this defense, I will present an overview of the research works I have supervised on Motion Capture and how it has been linked with Measurement and Modeling. The term of Motion Capture is more widely spread than Motion Measurement. From this fact, one might see the intuition that motion is a phenomenon more complex than the dataflow delivered by positional markers, the current standard for rigorous metrology. If a marker can be quantified, motion thus keeps a special quality which is interesting to characterize. To this end, my research work has been initially dedicated to 3D animation and has evolved to topics related to anatomy in motion. Contributions to 3D animation have been first on the search for optimal spaces to analyze and generate articulated motion. Optimality must be considered here in the sense of a dimensionality reduction which best preserves the quality of motion. This general approach has been applied to motion analysis from video, for humans and animals, compression of motion capture data and pose edition of 3D character. After this kinematical phase, another part of my work has been the integration of real data into physically-based animation. In this context, a physical simulation of a quadruped model based on PD-controllers has been optimized with respect to ground truth data. Following a different approach, human motion has been physically modeled with a Lagrangian method applied on the optimal spaces previously studied. To get closer to measurement while staying in dynamics, methods of prediction of forces from kinematics data only have also been studied in case of multiple contacts. Finally, a large part of my research work has also been dedicated to the development of experimental platform for laboratory animals, typically mice and rats. Tests on rodents are the first steps for any public release of medical drugs and are largely used in genetics for phenotyping. The motor activity is one of the most important components of behavior. To improve the automation of ethological methodology, I have developed platforms for 3D measurement of rodent motion from video, in various conditions, from anatomy and neurophysiology laboratories to low-gravity flights. 3D anatomical models have been developed and coupled with video-based motion analysis methods. This experimental research work has led to the set-up of a new analysis platform based on the coupling of multiple view video cameras and biplanar cineradiography, which opens now new perspective for anatomical motion capture.

President:

Mme Marie-Paule Cani (Professeur - Grenoble INP)

Raporteurs:

• Mr Ronan Boulic (Senior Researcher - EPFL )
• Mr Jean-Paul Laumond (Senior Researcher - CNRS-LAAS )
• Mr Karan Singh (Professor - Univesity of Toronto )

Examinators:

• Mr Michael Gleicher (Professor - University of Wisconsin Madison )
• Mme Sylvie Gibet (Professeur - Université de Bretagne )