Information Technology Reference
In-Depth Information
easing thus the segmentation process considerably [2, 13, 50, 51]. Postures are also
commands by themselves and many systems have been developed that use posture
recognition solely in order to transmit various commands [13, 24, 26, 49, 50, 60].
Vatavu et al. [51] use a specific hand posture in order to detect when motion com-
mands are entered: recording starts when the finger is pointed and ends when it has
been retracted. This gives users better control on how to enter gestures by employing
the same principle of click-based interaction. Figure 6 illustrates the segmentation
process aided by this specific posture that transforms thus into a motion command
detection event. The same click-like events are used by Vatavu and Pentiuc [50] for
a TV control application: gesture motions are being recorded and interpreted only
when predefined postures are detected (hand open and hand closed have different
associated recording actions). The idea is similar to that introduced by Freeman and
Weissman [13] but shifted towards a different interaction space that allows privacy,
interface sharing and, very important, robustness of the implementation.
Fig. 6 Motion gestures are recorded while the hand is in a predefined posture: the hand closed
with the index finger pointed out. The posture acts as an event trigger signaling both the start
and stop timestamps for the motion command the user is entering.
A common approach for detecting and tracking human hands in video is to use
color-based detectors that learn the distribution of the skin in a given color space.
Instead of wearing distinctly colored gloves or rings [54, 64] or using LEDs or any
other marker systems [48], skin color detection has proven to be an important inter-
mediate step for face and hands tracking [4, 8, 28]. Other approaches are to track
good features [47] or sets of such features as in the flock of birds HandVU imple-
mentation of Kolsch et al. [24, 26]. The Haar-like features that work so well with
face detectors have been equally investigated for detecting various hand postures
[25]. Also, more than one camera have been used in order to improve the accuracy
of the tracking process [9, 44].
Such advanced and complex tracking algorithms are not always needed for the
reliable acquisition of the human hand. For example, Wilson [60] introduced a sim-
ple solution for detecting hand gestures in order to simulate cursor functionality
 
Search WWH ::




Custom Search