|
Demos
1. Nouse basic
|
|
MPEG/AVI
videos - more here.
|
 |
Filename:
range-of-tracked-motion.mpeg
(1.6Mb) & speed-n-robustness.mpeg
(0.9Mb)
Description:
Nouse at work.
See
these
demos first
to see range
of motion and speed
& robustness of Nouse.
|
 |
Filename:
closeview-robustness.mpeg
(4Mb), overview-speed.mpg
(3Mb) & more
Description:
StereoTracker at work.
These
demos
show the range
of motion and the speed
& robustness of the 3D Nouse-based Stereo Face Tracking
|
 |
Filename:
Gclef.avi (260Kb)
& Gclef-ani.gif
Description:
NousePaint at work.
Test - The user rotates his head only! (the shoulders do not move)
|
|
Filename:
home-of-nouse-with-distractions.avi (1.2Mb)
Description:
NousePaint at work.
Test - The user writes a sentence with the nose, with his colleagues bothering him.
|
|
Filename:
ImmersiveEnvWNouse.mpg (4Mb)
Description:
Navigating in Virtual 3D Worlds with Nouse (in a mouse mode).
In this demo, the user rotates her head to follow a moving object (a car) in a
virtual 3D world. Nouse tracks the user's nose and adjusts the
first-person 3D view according to the rotation of the head: lifting the head up causes the
view to go higher and so on.
|
|
Filename:
PlayingWNouseBF.mpg (3Mb)
Description: Precise aiming with the nose (Nouse in a joystick mode)
A user plays a well-known BubbleFrenzy
game, the goal of which is to aim the turret to match the bubbles.
Traditionally played with a mouse or
key presses, it can now be played in more natural way by pointing the direction
with the nose.
Players say: "Playing the
game with Nouse is not only more fun, but is also
less tiring!" - Some users experienced
severe wrist fatigue when they played the game using a mouse for longer
than 15 minutes. This does not happen when they play it with with Nouse.
Using the nose
to aim the turret is found very natural, while the precision of
aiming with the nose was as good as with mouse. |
|
Image collection
You
can write or operate with Nouse as
with a joystick or a chalk.
|
 
|
|
You can
navigate in Windows environment
hands-free.
More drawings
made hands-free by nose are here.
|
|
|
|
2. Other
Perceptual Vision tools
2.1 Blink detection using second-order
change detection
|
Filename: bt-result.avi
(440Kb) & NousePP.avi (160Kb) &
more
Description: Blink
Detection uses
second-order
change detection to
detect eye blinking, switches Nouse On/Off by double-blinking.
See
also image below.
|
|
To see how second-order change detection helps
detecting blinks in moving heads, watch: Filename: ddI-blinking-face.avi
Description:
ddI-blinking-face.avi
Filename: floppy-ani.gif
Description: in
this example, of a floppy
disk moves from left
to
right with
a
protective
cover
sliding
from an
open to
close
position.
|
In the
bottom
row:
First
image (on
the left)
- this is
what you get
using
ordinary
(first
order)
change
detection.
Second and
third
images -
this is
what you
can remove
from the
first
image.
Right most
images -
this is
what you can
get to
detect
local (second-order) change.
|
2.2
Skin detection for face tracking and
augmented reality
Filename:
conducting-ani.gif,
4-4.avi, faces.avi
& more
Description:
We have examined many non-linear colour spaces
for
optimal skin model representation. The best result were
obtained, using Perceptual Uniform Colour
Space, which is the space that approximates
the colours the way humans perceive them.
Motion information allows to filter out the
spurious skin-looking regions
2.3 Face Detection and
Tracking with Multiple Cameras using
three-channel video representation
Can
you
find six different webcams on the picture at right?
(They all have different colour
adjustment properties)
|
All of them are
running at the same time,
detecting and tracking a
face:
To see
the result (with
lights off and on) view these images:
6cams-fd-lights-off.gif
6cams-fd-lights-on.gif
6cams-fd-lights-on1.gif
Note that switching on the lights is
not detected as a motion change - which is due to
the non-linear change detection.
Also note that, even when one channel
(or one camera) fails detecting a face, the others
in most cases do not.
This allows one to detect and track a face very
robustly (regardless of face orientation, etc)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2.4 On-line face memorization
and recognition for user registration and
identification
You can also see the animated
gif (600Kb) and AVI
movie (2Mb - watch in 50% size to see the
entire picture) which show the snapshots of
the program during the on-fly
memorization and retrieval of faces from
video and face database.
The sequence of actions in the movie is the
following:
- a face of the current user is
memorized,
- 62 faces are loaded from the face database
(using
face_data-BioID.txt face list -
first image is memorized used)
- recognition is performed on the same 62
persons shown with second image of each person
from the face_data-BioID.txt
list used
- video is back on and the user is again (and
still) recognized every time he blinks.
The log file with recognition statistics
for this run is given here.
The content of the memory (represented by the
memory synaptic matrix) shown in the right top
image. Grey image means nothing is stored -
all weights are zero. As more and more face
are stored the matrix approaches the identity
matrix. By analyzing this memory image, one
can always analytically estimate the
quality of face retrieval - more about the recognition
at this
website.
|
|
