Virtual reality gloves were originally designed as a replacement for the keyboard however it soon became apparent that its applications ranged from computer games and other virtual environments, and as a method for those who rely upon sign language to communicate with others.

Designs have ranged from a simple glove to a full exoskeleton worn over the arms and hand. While an exoskeleton undoubtedly provides a high resolution over the full range of arm and hand movement, it can be difficult to get into, and is not very portable. It can also interfere with some signs by the necessary bulk impeding movement and finger interaction. A glove does not suffer from the latter problem as much and is quite portable. The drawback here is a loss of resolution for determining movement and durability (the necessity of wearing and removing the glove can cause circuit faults quite easily).

The problem then comes down to a glove that can cover as much range of motion as possible while still retaining a low profile and high accuracy. Where some movement of the hand can not be identified then so that movement should not be one that sign language relies upon. In addition to these requirements the design should be cheap – note that that does not mean it is not complex, or that it should not be durable, but merely that the technology involved does not require large monetary funds to reproduce.

With the advent of digital image technology there has been much interest in sign language recognition by method of computer linked cameras (such as web cameras). Most devices today are based upon Charged-Couple Device (or CCD) technology (http://www.wordiq.com/definition/CCD). Such devices can obtain relatively high resolutions at a reasonable speed and are quite small and portable. Their major drawback is that the computing power to analyse each frame in real time and convert it to an appropriate sign simply does not exist in a portable format as of the time of writing. There is also the problem of dimensions; a single CCD camera only works in 2D space, where as sign language is 3D. Two or more cameras can help to overcome this problem, but complexity is enlarged substantially in this case. As a result, the more complex components required (generally finger movement) are relegated to a glove while elements such as hand position and orientation are left to the camera.

When designing a glove, an important note is the base requirements in determining hand motion made and recognised through sign language. Detecting the amount of finger splaying is redundant information is all that is required is that it be known if the fingers are splaying or not. This information is binary by nature, which simplifies design to a contact switch. As a side note, much of the equipment used in detecting finger bending can also be adapted to sense the amount of finger splaying – this is an important point as it means future directions of sign language can also be covered with little modification to the original glove design.

In determining a sensor to detect the amount of bend of a finger the first step was to study just how a finger moves. Pivotal axes exist on each knuckle (x-axis) with an additional degree of freedom for the base knuckle (z-axis). The important aspect about each axis of rotation is where its centre is. For x-axis rotation, the pivot is located in the middle of the finger. Anything laying on top of the finger will therefore stretch (if able to) as the total length across the finger is increased. This attribute is used by many flex sensor devices, but only in the capacity of vertical compression. The flex sensor, or angular displacement sensor, (2) is a commercially available sensor commonly used in virtual reality gloves to detect finger bending. The technology is based upon a conductive ink that is compressed, and so changes the resistance of the flex sensor depending upon compression. Bending said sensor causes this compression. Although one was not obtained for testing purposes, the ink can be modified for increased durability.
This technology appears to be standard in the virtual reality glove industry and so at first glance appears to be an appropriate answer, and is indeed so for most sensory glove configurations.. Several reasons make it unsuitable without modification however – primary reasons being the length and cost.  Each sensor is approximately 4.5” in length and so would not easily cover only one knuckle. It would also not be able to be moved along the glove – this causes problems as it would be expected that a wearer may have to customise the glove to suit their individual hand sizes. These drawbacks could be overcome with modification to the basic flex sensor design but the increased expenses as a result make other alternative more viable. Other companies do provide cheaper solutions however they are not designed for the durability required of long term usage in a sensory glove.

Other methods using similar attributes of compression upon finger bending were researched, including fibre optic sensor (abandoned due to the brittle nature of optical fibre strands) and capacitor plate distance. It was soon realised that vertical compression was not the only attribute that could be attached to finger bend motion; horizontal displacement could also be achieved. This is a much more viable alternative with the use of capacitor plates and also gives a high degree of accuracy. It facilitates more moving parts than other methods but a mechanical solution of some degree as the nature of the problem itself is machanical.

Capacitance acts as an open circuit where DC currents are concerned. But during the change from low to high voltage, this becomes a changing current until a voltage potential steady state condition is reached across the plates. When power is no longer connected to circuit the capacitor discharges. By measuring the voltage level a few moments after power is removed the amount of capacitance (which is in turn changed by the motion of bending a finger) can be determined. In practice this is not actually done, but a linear scale of relative values to amount of capacitor plate displacement can be determined, and so numbers on the relative scale can be mapped to finger motion directly. For this to work the RC constant of the circuit in place must be appropriate such that time taken for the capacitor to discharge be long enough that a microchip analogue to digital converter (AD converter) can take a reading, yet not so long that overall response of the glove is noticeably slowed to users as a result (bearing in mind that as many as 22 connected capacitors may be present).

Calculations made with the various materials available placed average capacitance (full plate overlap) at approximately 20pF. This assumes a dielectric value of 1.5 and plate separation distance of 1mm. With this in mind, resistor values of one meg-ohm and 10 meg-ohms were chosen and tested. The 1 meg-ohm resistor was chosen for several reasons; the results were more than adequate and the charge time for the capacitor was much shorter than with the 10 meg-ohm (it must be remembered that while the RC represents just how long it takes to discharge, the capacitor must also charge and this time is wanted to be as small as possible).

After a brief test with the m16c microprocessor and the built in AD-converter it was shown that a 10degree accuracy was possible with large error margins – more degrees of accuracy were also possible but depended entirely upon the stability of the capacitor housing (for example, if the plates could move such that the distance between them changed, then this affects the results dramatically). Production standards would decrease the error rates present here and could feasibly provide 5 degree accuracy.
For this test plastic sheet (from the front of a loose leaf binder) was used for the dielectric (approximate value of 2) with copper sheet for the plates. The plastic sheet was chosen as it is a widely available material, offers a relatively high dielectric and is thin. A touch of superglue was used to attach rectangular cut elements to one capacitor plate. The opposite plate is then free to slide over the plastic as appropriate to the design. Copper plates were chosen as they provide a stiff conductive medium that will not bend easily with movement from the hand. If the plates were to change their shape this could not only affect the ability of it to slide easily, but also the distance between plates and surface area overlap, thus producing errors in readings. Copper was also used for the ease with which connections are soldered onto the plates.

When soldering copper plates care should be taken that the heat from the soldering iron does not warp metal. As with any bending of the plates this could produce inconsistency in the results or problems with the sliding action.

Through all of this it must still be remembered that it takes time to send the information back through the microchip to the attached PC via a serial cable. At an estimated 4ms round trip time. Given that a sensory glove is a real time device, the maximum time between sensor readings, processing and transferral to the PC (which given today's technology can be assumed to be negligible time) of 10ms. Much higher and the degree of accuracy with which signs can be interpreted decreases substantially (especially for those signs that are completed swiftly). Proper program design and implementation should keep this goal well within the limits of the microchip (it should be able to produce a full hand sensor sample in much less time than required).

2)
http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=/netahtml/search-bool.html&r=1&f=G&l=50&co1=AND&d=ptxt&s1=5086785.WKU.&OS=PN/5086785&RS=PN/5086785 	- flex sensor patent address.