Résumé des résultats (Abstract)
(Anglais)
|
EPFL's 2001 achievements:
Completion of Body Animation Decoder Integration: Body nodes have been finalized and integrated into the player as well as realtime FBA decoding functionalities. The offline scenario where a MP4 stream containing both body models and animation data is properly decoded and displayed is ready. The interactive streaming of on-demand body animations into a preloaded scene has also been successfully tested.
FBA Encoding tool and library : A simple graphical user interface allows content providers to create a single compressed binary output using the files generated by the Face Animation plug-in and the Body Animation Production plug-in. It is based on a realtime library which uses techniques such as quantization, progressive encoding and arithmetic encoding. In addition to a .fba file which contains the compressed FAPs and BAPs., the tool creates a .hint file providing additional information about every FBA unit, which makes possible the encoding of the FBA data into the MP4 stream using MP4Tool. The library is also provided separately and has been successfully integrated into the AML processor.
Library of BAPs : EPFL has provided a set of predefined low-level body animation files (.bap) that can be streamed to the player using the AML processor. Those animations have been manually created into 3DS Max and exported using the MPEG4 body animation production plug-in or have been produced using magnetic motion tracking hardware. They have been selected to satisfy behaviour specifications, especially improving communication in a chat application context by allowing nonverbal messages to be sent, and are well suited to sales assistants and personal assistants characters.
High-level behavior library : In addition to the predefined animations, EPFL has delivered a binary library generating on-the-fly parametric animations that are useful in many scenarios but that can't be pre-calculated: facing, pointing and walking motion. When an high-level behaviour is called through the AML syntax, it receives a set of coordinates in the 3D world and generates the corresponding BAPs, taking into account the former position and orientation of the avatar in the scene, as well as providing smooth transition from the preceding posture if applicable. Such techniques as inverse kinematics or quaternion interpolation are used. This library is linked with to the BASParser library.
BASParser library : This library is able to parse the body component of the AML (Avatar mark-up Language) syntax, process the referenced low-level animations (modify speed, intensity or extent, mix several concurrent animations using a priority parameter, etc.), and output low-level BAPs according to given timing information. Since the duration of parametric animations may not be known beforehand, body animations can also be triggered using timing information relative to each other (e.g. simultaneous or temporally serial). This library is part of the AML processor.
|
