Much of the discussion about instrumented gloves comes from the excellent review by Sturman and Zeltzer [SZ94].
The first ``generation'' of gloves were not intended for use for gesture recognition. This included the Sayre Glove, a glove which was really intended for control of as many devices as possible through the manipulation the hand.
This is the point where people began to consider the alternative of analysing video images, and see if this was a practical way of analysis.
The first widely recognised device for measuring hand positions was developed by Dr G Grimes at (surprise, surprise) AT & T Bell Labs [Gri83]. Patented in 1983, Grimes' ``Digital Data Entry Glove''. It had finger flex sensors, tactile sensors at the fingertips, orientation sensing and wrist-positioning sensors. The positions of the sensors themselves were changeable. It was intended for creating ``alpha-numeric'' characters by examining hand positions. It was intended as an alternative to keyboards, but it also proved to be effective as a tool for allowing non-vocal users to ``finger-spell'' words using such a system.
Figure 2.6: A picture of the internals of the VPL
DataGlove, attached to a Macintosh.
This was soon followed by a more ``regular'' device, which was later to become the VPL DataGlove [ZL87,ZL92], which is shown in figure 2.6. This device (which is, by the way, heavily protected by patent laws) was built by Thomas Zimmerman, who also patented the optical flex sensors used by the gloves. These sensors have fibre optic cables with a light at one end, and a photodiode at the other. Zimmerman had also built the Z-glove which he had attached to his Commodore 64, which was a simplified version. This device measured the angles of each of the first two knuckles of the fingers using the fibre optic devices, and was usually combined with a Polhemus tracking device. Some also had abduction measurements.
This was really the first commercially available glove. Even so, at in excess of US$9000 a piece, they were not quite within everyone's reach.
Figure 2.7: The Exos Dextrous Hand Master
A completely new alternative was proposed at this stage, which was to become the Exos Dextrous Hand Master. This device makes no pretence of being a glove. Imagine an exoskeleton for your hand bedecked with sensors --- you will have a pretty good image of the DHM (it is shown in figure 2.7). By all accounts, it takes a while to get on, but in terms of accuracy of information, it measures 20 degrees of freedom with 8 bits of accuracy, at up to 200 Hz, which is currently unrivalled. Usually a 6D tracker is attached, giving a total of 26 degrees of freedom. While ideal for tele-robotics (remote control of robots), it is overkill for some applications. It is also not cheap (around US$15000).
Through a deal with VPL, Mattel/Nintendo began manufacturing
PowerGloves. It has many design corners cut. These include only having
sensors on the first four fingers, and a very poor tracking mechanism
(4D - x, y, z, roll vs other gloves' 6D - x, y, z, yaw, pitch, roll).
To quote Sturman and Zeltzer ``Although the least accurate of the
whole-hand input devices, the PowerGlove is also the cheapest by a
factor of 100.''
In addition, interface boxes
exist that allows the easy connection to almost any machine through a
standard RS-232 serial port (such as UIUC's PowerGlove Serial
Interface, or the Menelli Box).
In parallel with this James Kramer at Stanford University was developing the CyberGlove -- which is patented [Kra91]. This glove was specifically designed for use in his ``Talking Glove'' project, which is a system for communication between non-vocal people and speaking people. It comes in two models: one with 18 degrees of freedom and the other with 22 degrees of freedom and measures bend of the first two knuckles on each finger (three for the 22-DOF model), abduction between fingers and a number of additional measures around the thumb (since it has 5 degrees of freedom), by using strain gauges arranged in Wheatstone Bridge configuration. Together with a Polhemus tracker this would form the perfect system for this thesis. However, the cost is approximately US$6000 for the glove, not including the position-tracker.
As we speak, prices of these devices rapidly falling, as are the
tracking devices themselves. In June 1995, Fifth Dimension
Technologies released the 5th Glove '95, costing US$300,
although it only had finger bend measures and no abduction or
individual thumb measures. Also, in mid-September 1995 Abrams-Gentile
Entertainment (AGE Inc) announced the release of the PC-PowerGlove,
which was to have 8-bit information on each finger, 16-bit position
information, 8-bit orientation information (roll, pitch, yaw) and also
the possibility of a pair of gloves, which would allow much
improved recognition, for the surprising price of US$120 each. The
techniques developed in this thesis could be mapped to these new
gloves in a matter of days.
In addition, high-accuracy magnetic trackers are also falling in price, with both Polhemus and Ascension trackers now available in the sub-US$1000 region.
So while currently the technology is not within the reach of many, this is changing rapidly. While virtual reality may not be growing at the rate originally predicted, it is still growing at a sufficient rate to see the prices on these devices fall quickly.