Real time, multi-level
motion analysis (part 2)
Motion analysis in real time, as part of an
interactive composition, creates increased challenges
- the
motion capture challenges described above decrease
substantially the quality of marker labeling, particularly when
multiple movers are involved
- real time constraints limit the amount
of computations that can be performed before an analysis decision
needs to be made and feedback
provided
- the computational load sometimes exceeds the ability
of even high-end computational resources available at labs
like AME/IREMA
- no opportunity exists to look forward in time
for testing and corroborating analysis decisions
Real time motion
analysis accomplishments of the motione project
Three key accomplishments have been developed and implemented in
real time motion analysis for the motione project by the AME/IREMA
team:
- real time pose recognition engine that works with
labeled and unlabeled markers
- engine using only unlabelled markers
that: tracks multiple dancers, recognizes dancer groupings
and shapes in global space, calculate dancers’ speed of
movement and rate of activity, calculates translational velocity
correlation
and activity correlation of dancers and related offsets.
- engine
for the extraction of statistical (probabilistic) middle
and high level features of choreography (form extraction) and
use
of these features for improvement of low level analysis.
|
