|
The KidsRoom re-created a
child's bedroom. The space was 24 x 18 feet, with
a wire-grid ceiling about 25 feet high. Two of
the bedroom walls resembled the real walls in a
children's room, complete with real furniture,
posters, etc. The other two walls were large
video projection screens used to
"transform" the room between different
storybook worlds. Images were back-projected onto
the screens from outside of the room. Behind the
screens was a computer cluster with six machines
that controlled the room. Computer-controlled
theater lighting on the ceiling grid lit the
space from above with white light, and several
colored lights were used for the transition
scenes. A microphone was paced inside the room to
detect the volume of screams. Four speakers and
one amplifier projected sound effects and music
into the space. Finally, there were four video
cameras in the space. Three were used for
computer vision, to recognize the actions of the
children, and one was used for spectators to view
the room when people were using it. |
|
Camera Views
Three cameras were used for the
computer vision, and one for spectators.
Camera 1:
Top View
Used for tracking people in all worlds
and for detecting rowing in the river
world. |
|
|
Camera 3:
Red Rug
Used for action recognition during the
monster dance. |
Camera 2:
Green Rug
Used for action recognition during the
monster dance. |
|
|
Camera 4:
Spectator View
This view shows both the red and green
rugs and partial views of both screens
for spectators. |
Each camera view was
selected for a reason. Camera one was used by
the object tracker and is a convenient view
because the entire room is visible and there
is no occlusion each object is always
visible in the scene and not blocked from
view by other objects. Cameras two and three
were used during the MonsterLand world to
recognize actions on the red and green rugs.
These two camera views were selected so that
two, non-occluded views of each person on the
red or green rug could be obtained, one from
the back and one from the side. The action
recognition
technique used by the KidsRoom can use
multiple viewpoints of the same action to
make recognition more robust, although in the
actual KidsRoom implementation only one view
was used for each move. The final camera was
placed so that spectators outside the room
could see most of the room in addition to the
two displays. |
|
Room Layout
|
The room contained several pieces of real
furniture. The main object was a moveable bed,
which was used throughout the story. The other
furniture, which was not explicitly tracked by
the computer system, was fastened to the ground
and nailed shut. The rugs on the floor, also
fastened carefully, provided points of reference
during the story so that children were standing
in locations where the cameras could see them. A
path of simulated stones was marked on the floor.
Four speakers in the space delivered music and
sound effects. One speaker was on each wall, and
many sounds were directional. The microphone was
located between the two large screens. The
entrance way, in the bottom left corner of the
room, was monitored by the computer so that
people could enter and exit at any time when the
room was active. Cinder blocks on the floor near
the screen prevented enthusiastic children from
pushing the bed through the screens. |
|
|
Equipment
|
The six computers were:
1 SGI Indy R500
workstation
Used for tracking, playing sound effects,
MIDI light output
(It might be possible to replace
these workstations machines with PCs with
special video hardware.)
1 SGI Indy R5000
workstation
Used for action recognition on one rug,
energy detection in river scene,
sending MIDI music commands to the
Macintosh, amplitude audio detection.
1 SGI Indy R5000
workstation
Used for action recognition on the second
rug
2 Digital AlphaStations
Used for displaying animations, one per
screen
(These could be easily replaced with
PCs.)
1 Macintosh
Used for running Studio Pro MIDI
software.
(Midi control could be done on a PC.)
Other equipment:
2 high-resolution video
projectors and wall-sized screens
4 Sony HandyCam color
video cameras (3 for vision, 1 for
spectator view)
4 speakers and a
12-channel 4-output mixer and amplifier
1 Radio Shack PZM
microphone
14 lights (11 white, 3
colored), controlled by a MIDI-based
light board
Local network connection
Misc. cables and video
distribution amplifiers
|
|
|
|