CALL FOR PAPERS
AAAI Workshop on Perceiving and Interpreting Action
http://www.media.mit.edu/pia97/

At the National Conference on Artificial Intelligence,
July 27-31 1997, Providence, Rhode Island, USA


Recent advances in technology and economy of computer vision have created
wide interest in interpreting action, particularly that of people.
Anticipating a day when real-time vision becomes a significant medium for
human-computer interaction, many researchers have proposed inference-rich
applications: Virtual assistants that help mechanics do repairs; digital
coaches for dancers and athletes; vision-driven VR applications; safety
monitors that look for trouble in baby-rooms, factory floors, and traffic.
Some parts of these applications have already been prototyped. However,
connecting perception to inference and determining what inferences should
happen remain looming problems.
Efforts toward action-understanding may require or spur advances in
non-rigid motion tracking, event perception, visual learning, probabilistic
inference, causal and temporal reasoning, plan recognition, and models of
intentionality. The workshop is aimed at bringing together researchers in
perception, AI, learning, and psychology whose work connects with the
perception and interpretation of action. We expect to see papers on topics
such as:

    Visual representations for motion interpretation.
    Motion pattern classification for articulating bodies.
    The spatiotemporal structure of actions.
    Interpreting gestures in context.
    Inferring context (tasks and activities) from video/audio/proprioception.
    Temporal inference over approximate and noisy data.
    Learning and recognizing procedures from video.
    High-level models of action and intention.
    Inferring plans and goals from sensing via high-level models.
    Systems capable of sustained human-computer cooperation.
    Other perceptual modalities.

The 1-day workshop will consist of four topical sessions of research
presentations, each led by an invited talk or tutorial and capped with a
short panel discussion. Attendance will be limited to 30 people to
encourage group discussion. In addition to working notes, we expect to
produce a digital proceedings which can be browsed over the web after the
workshop.

Submissions

Interested researchers are invited to submit short but complete technical
papers (up to 8 pages, 4000 words) or statements of interest describing
relevant research. We are interested in both mature research and early
results from works-in-progress.  Electronic submissions are strongly
encouraged: ascii, postscript, or html (self-contained directories packaged
via tar, Stuffit, or pkzip) should be deposited via anonymous ftp in
ftp://pia97.media.mit.edu/incoming and an email message should be sent to
pia97@media.mit.edu containing your name, title,
abstract, ftp file name, and mail/email/WWW addresses.  WWW pages with
step-by-step
instructions and a submission form can be found at
http://www.media.mit.edu/pia97/

Hardcopy should be sent to

    PIA97 c/o Matthew Brand
    MIT Media Lab E15-385
    20 Ames Street
    Cambridge, MA 02139-4307
    email: pia97@media.mit.edu
    phone: 617.253.0608
    fax: 617.253.8874

Key Dates
    11mar97 Submissions deadline
    1apr97 Notifications
    22apr97 Camera-ready copy deadline
    27/28jul97 Workshop

In the event that many intriguing papers are received, the workshop may be
expanded to two days.

Organizing Committee

Aaron Bobick, Matthew Brand (chair), Sandy Pentland, MIT Media Lab,
{bobick,brand,sandy}@media.mit.edu; Stan Rosenschein, Stanford/Autodesk,
stan.rosenschein@autodesk.com; Michael Swain, University of Chicago,
swain@cs.uchicago.edu