Reactive Virtual Trainer (RVT) is an Embodied Conversational Agent
capable of presenting physical exercises that are to be performed by
a
human, monitoring the user and providing feedback at different
levels. ECA, parameterized animation, fitness, multimodal coordination Some themes present in this work are :
|
To find out how tempo affects motion dynamics, we analyze optical
motion capture recordings of subjects performing excercises.
|
|
This work is done in a Bachelor Referaat assignment by Frank Zijlstra.
The Virtual Trainer is likely to be used in a home entertainment
environment. Only one webcam will be available and lab conditions
should not be required for the analysis to work. To provide feedback
timely, the analysis should work in real time.
|
Using a silhouette based approach, these requirements can be met. The
silhouette is extracted from a video recording. In this silhouette,
body parts are identified in real time.
|
|
|
The body part are matched to body parts in a reference silhouette to
provide feedback on how to correct the pose.
|
|
|
We make use of the multi-modal synchronization language
BML to specify the
synchronization between
metronome beats, speech and clapping motion.
|
|
By playing the users preferred music during the exercise, exercising
could be made more enjoyable. We interpret the music to find the beats
so that the exercise, and thus the movement of the trainer, can be
aligned to the tempo of the music. For this we can borrow the beat
detection algorithm from our Virtual
Conductor, which implements methods described in the
publications of Anssi Klapuri. This algorithm will detect the tempo
and beat in the music played during the exercise.
Motion captured fitness exercises can be found at the HMI mocap database.
Using our animation framework, we can define exercises as by
describing movement paths of limbs. Movements can be accentuated by
adapting amplitude or velocity profile.
Parameterized animation (1Mb).
|
|
Former HMI-members:- GATE [GATE: Games for Advanced Training and Entertainment]
- HUMAINE [Human-Machine Interaction Network on Emotion]
|