Shahzad Malik, Joe
Laszlo
Department of
Computer Science
University of
Toronto
hand segmentation
Using
background subtraction, the hand regions and contours can be extracted from
the image. Strong peaks in the contours
are then marked as fingers, with a heuristic approach to decide on finger
labels. Finger orientation is
determined by fitting a line through the fingertip and the midpoints along the
finger region. Postures and gestures
can then be recognized by using both static and temporal finger
information.
t e
c h n o l o g y
touchpad
detection
Assuming the touchpad is a simple piece of paper with a black rectangle
defining the active area, image thresholding is used to extract the touchpad
from an image. Using a homography, the
corners of the touchpad are then mapped to the corners of the display.
o v
e r v i e w
a
low-cost, multi-point, touch-sensitive device
By using two off-the-shelf web cameras and simple computer vision
algorithms, the Visual Touchpad provides a low-cost alternative to more
expensive touch-sensitive hardware. The
system can detect fingertip positions for all five fingers of each hand, as
well as the 2D orientation of each finger.
layered
gesture recognition
Unlike standard touch-sensitive surfaces, the Visual Touchpad allows
hand gestures to be recognized both on the touch surface, as well as above it,
allowing for a wider range of gestures and interactions.
“hands-up”
display
By segmenting the hands from the live video streams, they can be
rendered directly on the screen. This
allows for a compelling direct manipulation interface that compares favorably
to touch-sensitive screens or surfaces, but without the occlusion or fatigue
problems associated with such devices.
single
device for text-entry and navigation
With the ability to detect all five fingers of each hand, the Visual
Touchpad can be used for natural text entry with a customized, on-screen
virtual keyboard. This can potentially
replace both the keyboard and mouse with a single interface, which is useful
for applications that require a user to frequently alternate between
text-entry and mouse navigation modes.