Human Motion Capture and movement modeling

Abstract

We propose a new technique for the animation of virtual actors based on the magnetic sensors and datagloves VR technologies.
Two kinds of applications are presented, first type are real-time applications using motion capture to replicate the movement of the participant (e.g. collaborative interface to shared virtual environment), second domain is the recording of qualitative human gestures and movements demanding more computation efforts.

Summary of the project

The purpose of this project is to provide and constructs new animation methods based on the currently available VR hardware interfaces. We have achieved an efficient method for converting magnetic sensor data in human anatomical angles. Our method comprises three stages: the skeleton calibration, the automatic sensor calibration, and the final conversion step. The skeleton calibration consists in adjusting the virtual skeleton to the stature and segment lengths of the performer. The second stage, automatic sensor calibration, is an efficient initialization technique constructing the relation between the sensors and the virtual joints of the virtual human hierarchy. Once the sensor calibration has been performed, the anatomical angles used to animate the virtual human can be evaluated in interactive applications or recorded at high rate to produce key-frame files.
Several improvements of the basic converter algorithm are designed, such as knee and elbow twisting specific evaluation technique, floor contact constraint enforcement, and multi-joint control of the spine, shoulder-clavicula joints and non-Eulerian ankle model.

The animation of the hands is achieved using different equipment and technology. Consequently, specific algorithms are constructed for the knuckles management using the dataglove parameters. The angle values measured by the dataglove devices with the help of some associated hand model are mapped onto our own human hand model.

Client applications for this work are illustrated by several prototypes among which Virtual Karate (L. Emering) an AgentLib demonstrator of real/virtual actor interaction, the latest version of Track-Motion Capture (J. Huang), fobnet (T. Molet) a human driver for VLNet (T. Capin, I. Pandzic and L. Elwin), and birdy (T. Molet) the evolution of the Scenelib modeler in the Motion Capture and interactive grasping (S. Rezzonico) context.

Some human motion capture results have been demonstrated at the Genevaís Telecom í97 through the virtual tennis and the virtual dancer public demonstrations.

Publications

T. Molet, R. Boulic, D.Thalmann, A Real-Time Anatomical Converter for Human Motion Capture, Proc. 7h Eurographics Workshop on Animation and Simulation, Springer-Verlag, Wien, September 1996.

Molet T., Huang Z., Boulic R.,Thalmann, D. An Animation Interface Designed for Motion Capture , Proc. Computer Animation'97, IEEE
CS Press, 1997, pp.77-85.