Modeling and analyzing human individual

1
Modeling and analyzing human individual and
collective spatial behavior Robert Sekuler, Maja
Mataric, Kristina Lerman Dylan Shell Brandeis
University, University of Southern California,
and USC Information Sciences Institute
BACKGROUND
THE PHYSICAL MUSEUM
THE VIRTUAL MUSEUM
The project brings together researchers from
computational science, mathematical modeling,
and cognitive science in order to study
individual and collective use of space. We seek
to understand human individual and collective
behavior on various temporal and spatial scales.
Our research focuses on a rich indoor
environment, the California Science Center (CSC),
where the behavior of individuals and crowds can
be observed analyzed, modeled, and simulated.
Laser-based people tracking. A pair of laser
range-finders are deployed in the museums Cell
Theater and Life Tunnel exhibits. Each device
independently logs to a portable workstation.
Data are run through a background-learning and
foreground-separation algorithm. A foreground
region of sufficient size is counted as a person
within the scene, and results are benchmarked
against observations by human judges. Video-based
people tracking. Low-resolution images from
cameras overlooking the museums open spaces are
automatically processed to detect people within
the video frames. Our algorithm looks for
circular regions of high-contrast that differ
from the computers current model of the image
background and slow changes in the image arising
from lighting changes. The goal is to calculate a
spatial density function over the monitored
space. To protect privacy, neither form of
tracking captures the identity of any individual
visitor.
The virtual museum implements the CSCs main
floor. The virtual museum is populated by
varying number of virtual visitors (bots),
whose spatial distribution and actions can be
controlled. Human subjects receive a set of goals
--such as locate a magnet in the gift shop--
and a limited time to achieve those goals. The
time limits and goals are based on observations
of visitor behavior in the CSC. The buildup of a
visitors cognitive map of the museum is
reflected in the changing behavior of subjects
who are initially unfamiliar with the museum.
Run-time software, now in test trials, captures
--for offline analysis- a players position and
movements, along with the movements and spatial
distribution of all bots.
TECHNOLOGIES

• Our project integrates data generated by three
  different, complementary technologies
• Physical Museum. Sensing of positional and
  movement information for individuals and crowds.
  The positions, movements and spatial distribution
  of museum visitors are measured using laser-based
  and video-based people tracking. The laser-based
  system works well in circumscribed spaces, like
  those for special exhibits the video-based
  system is best in large open spaces, like those
  in the museums main halls.
• Modeled Museum. Mathematical modeling of
  collective dynamics in virtual and real
  environments. The model describes visitors flow
  and transitions from one exhibit to another. The
  model generates predictions of flow under various
  conditions of physical obstacles and congestion
  by crowds. These predictions are compared to data
  from actual visitors to the CSC
• Virtual Museum. Goal-directed navigation in
  populated, interactive virtual environments. An
  interactive, virtual museum was built from floor
  plans and photographs of the CSC. Human subjects
  navigate the virtual museum to accomplish a
  series of goals, with varying crowd densities,
  time pressures and familiarity with the
  environment.
• Data from each technology will be used
  interactively, to guide future studies using the
  other technologies.

Wireframe view with bots paths (in blue)
THE MODELED MUSEUM
Model of spatio-temporal changes in distribution
of visitors. We are constructing and testing
probabilistic, macroscopic models of visitors
movements. The model focuses on ensemble
properties, such as flow. The model takes into
account exhibit capacity and viewing time, and is
evaluated against observations in the museums
Cell Theater and Life Tunnel special exhibits.
A video walkthrough the virtual museum is
available at www.brandeis.edu/sekuler/vMuseum.
ACKNOWLEDGMENTS
Unidirectional flow through exhibit stations
(from source to sink)
Support from the NSF Human Spatial Dynamics
program to the Principal Investigators is
gratefully acknowledged. That support,
supplemented by funds from NSFs Research
Experience for Undergraduates (REU) program, has
made possible the participation of three
undergraduate students and four graduate students
at the collaborative projects three sites. We
also acknowledge the gracious cooperation of the
CSCs staff. The virtual museum was constructed
by Digital Artefacts of Iowa City using various
tools, including real-time human simulation
software from Boston Dynamics. These tools were
chosen in order to facilitate the participation
of student researchers who are not expert
programmers. E-mail for the Principal
Investigators sekuler_at_brandeis.edu,
mataric_at_usc.edu, and lerman_at_isi.edu.
Bottleneck conditions Influx10 Capacities
c1c240, c310
Physical Museum
Bidirectional flow through exhibit stations (two
sources)
Measurements of visitors interactions and
activity
Modeled Museum
Virtual Museum
The model is parameterized by Rate of Influx and
Rate of Transition between exhibits. In turn,
Rate of Transition depends upon the capacity of
the next exhibit and the viewing time for the
current exhibit. The models present
implementation simplifies by assuming that
visitors are Markovian. Future extensions will
relax this assumption, allowing visitors
behavior to reflect longer histories, as well as
beliefs about likely conditions of crowding. We
will use the model in order to characterize
environmental conditions that promote crowding
and bottlenecks, and, in the future, identify
means by which crowding and bottlenecks can be
reduced.
Mathematical description of activity and
interactions
Activity and interactions incontrolled VR
scenarios