Recently, several devices for improving quality and efficiency human-c
omputer interaction have been proposed. This is mainly due to the requ
irements of virtual reality applications, where the greatest part of t
he most circulating devices must be worn by the user-e.g. gloves for h
and-tracking, complete or partial tracksuits. These techniques allow t
he estimation of most human body movements. On the other hand, the ado
ption of such devices is unsuitable for many applications since they a
re too constrictive-e.g. teleconferencing. For this reason, other simp
ler devices such as 3-D mice have been suggested. Most of the 3-D mice
proposed in the literature compel the user to be connected to the dev
ice and their behaviour is much closer to a Joystick rather than to a
true 3-D mouse. More recently, vision-based hand tracking systems have
been presented, but most of them constrain the user to wear gloves, b
racelets or to mark some points on the hand. In this paper, a vision-b
ased, glove- and mark-free hand tracking system is proposed. This is b
ased on stereo vision and can be implemented with low-cost architectur
es owing to its low computational complexity. In addition, the estimat
ion process defined is very robust and allows the simultaneous hand-po
sition tracking and the recognition of some hand postures, thus implem
enting a true non-constrictive 3-D mouse. (C) 1995 Academic Press Limi
ted