Along with the head-mounted display, the instrumented glove has become the publicly recognised face of virtual reality. The majority of media articles on VR feature one of these two devices, and books such as Rheingold (1991) and Aukstakalnis and Blatner (1993) feature a glove prominently in their cover artwork. It is interesting, given this close connection between VR and instrumented gloves, that such devices generally were not originally developed for use in VR.
"Sayre" glove
Although research had been conducted into heavy exoskeletons which strapped onto the hand as far back as the 1950s, the first of the modern lightweight hand-measuring devices was probably that developed by DeFanti and Sandin (1977). Their glove was based on an idea by Rich Sayre from the University of Chicago, and consisted of flexible light-conducting tubes mounted along the fingers of a glove with a light source at one end of the tube and a photocell at the other. The amount of light reaching each photocell was reduced as the corresponding finger was flexed, and the resultant changes in voltage produced by the photocells could be monitored to measure the bend of the finger (Sturman 1992).
Digital Data Entry Glove
Another early glove was that patented by Gary J. Grimes in 1983. Grimes designed a glove capable of converting hand positions into alpha-numeric characters. This glove used a wide variety of different sensors to measure the flex of finger and wrist joints, and also to detect contact between various parts of the hand such as finger tips. These sensors were positioned specifically to recognise the handshapes used in the single-handed manual alphabet, and hence this glove could not be used as a general-purpose hand- measurement device (Grimes 1983, Sturman 1992, Bevan 1995).
VPL DataGlove
Grimes' employer, Bell Telephone Laboratories, did not develop his glove further however, and so the first commercially produced instrumented glove was the DataGlove manufactured by VPL. This product arose out of the desire of Thomas Zimmerman to use his hand to control a musical instrument without touching it – in effect to actually play 'air guitar'. Zimmerman's initial prototype was built out of an old work glove and several flexible hollow tubes which conducted light. These tubes were coupled with a light source and photosensor and used to measure finger flexion in the same manner as in DeFanti and Sandin's earlier glove.
Zimmerman patented this device in 1982, and later went on to found VPL along with VR guru Jaron Lanier. VPL's researchers developed Zimmerman's early glove into the commercial version of the DataGlove, replacing the light- conducting tubes with more accurate bundles of fibre-optic cables. These cables were treated by being carefully scored or cut. As a finger and its corresponding cable are bent, the gaps formed by these cuts becomes larger, and therefore more light is lost from the cable and fails to reach the photosensor.
PowerGlove
In 1989 Mattell began marketing the PowerGlove as an interface device for Nintendo game systems. Jointly developed by VPL and Abrams-Gentile Entertainment, this device offered hand-measuring capabilities at a greatly reduced cost (around 1/100 the cost of the DataGlove). The glove itself was manufactured from heavy-duty plastic and the expensive fibre-optic sensors were replaced by strips of plastic coated in electrically conductive ink. The resistance offered by these strips alters as the fingers are bent. The cost reduction came at the expense of vastly lower performance – the PowerGlove measures only the overall flex of the thumb, index, middle and ring fingers
with only two bits per measurement. This makes the PowerGlove too limited for high-end applications, but its affordability has made it extremely popular amongst 'home-brew' virtual reality developers.
CyberGlove
The CyberGlove was developed by Jim Kramer at Stanford University as part of his research into sign-language recognition (the recognition aspect of Kramer's work is discussed in Chapter 4). Kramer found that the DataGlove was not sufficiently accurate for his needs, and initially developed the CyberGlove for his own use. It is now produced commercially by Virtual Technologies.
The major difference between the CyberGlove and the other gloves already described lies in the sensing technology used. The sensors in the CyberGlove consist of two flexible strain gauges mounted back to back. A pair of gauges is sewn into the glove over each joint to be measured and wired in a Wheatstone bridge configuration. The resistance of the gauges varies as they are bent and by sampling this variation the angle of the corresponding joint can be calculated.
The advantage of this style of sensor over the fibre-optic approach used in the DataGlove is that the variation in response of the strain gauges is linear over the entire angular range, whilst the DataGlove sensors have non-linear variation. This makes the CyberGlove's sensors more consistently accurate (Kramer and Leifer 1987, Virtual Technologies 1992).
The 5th Glove
The 5th Glove was developed by Fifth Dimension Technologies and was released in the United States in May 1995 by General Reality Company. Exact details of the gloves specification are unclear at time of writing, but it appears to be based on a fibre-optic sensing system similar to that used by the DataGlove. From the values quoted for sensor update rate (125 Hz) and complete-hand update rate (25 Hz) it can be calculated that the 5th Glove provides only a single sensor per finger (similar to the PowerGlove). However the sensing technology used is considerably more accurate than that of the PowerGlove.
The most interesting aspect of this glove is that at a cost of around US$500 it fills a gap in the market between the low cost and low performance of the
PowerGlove and the high cost, high performance option of the DataGlove or CyberGlove (General Reality Company 1995).
Dextrous HandMaster
Although it is worn over the hand, the Dextrous HandMaster (DHM) produced by Exos differs from the other devices described in that it consists of an exoskeleton rather than a glove. The exoskeleton is held to the user's hand by velcro straps. Each joint of the exoskeleton is equipped with a Hall effect sensor. This sensor is based on altering the magnetic field surrounding a semiconductor as the joint bends, which affects the voltage output by the semiconductor.
Due to the precise mechanical linkages between the movements of the hand and exoskeleton joints and the accuracy of the Hall effect sensors, the DHM provides a much higher level of precision than any of the other hand-sensing devices discussed previously. Error rates in the region of 0.5°-1° are attainable. As a disadvantage however this system is bulky and heavy when compared to glove-based devices. For these reasons the main applications of the DHM have been for robotic control systems, where an extremely high level of precision is required.
3.1.2 Camera based joint measurement
An alternative approach to specialised glove-based hardware is to use standard video camera equipment to capture a visual image of the user, and then make use of computer vision techniques to extract data about the hands from this image. This avoids the expense of using special purpose hand- measuring devices, but introduces other problems due to its indirect nature. Primary amongst these difficulties is locating the position of the hand and fingers within the image. The hand can usually be located relatively easily, particularly if user's clothing and the background are chosen so as to contrast with the colour of the hands. Extracting the position of individual fingers from within the image of the hand is a much more difficult task, as in many hand positions the fingers will be occluded by each other and by other parts of the hand. In addition the lack of colour contrast between the fingers and hand makes the digits difficult to locate even when they are fully visible. The most common response to these problems is to place artificial markers on the points of interest on the hand (usually the fingertips and the joints of the fingers). Placing multiple markers at different angles on the hand can partially help to overcome problems of occlusion.
Sturman (1992) states that LEDs were commonly placed as markers on the hands, body or limbs of subjects in biomechanical research during the 1980s. The light emitted from these devices makes pin-pointing them in the camera image easier, particularly if filmed against a dark background. Other researchers such as Holden (1993, 1995) and Dorner (1994) have used specially coloured gloves worn by the user to aid in this feature extraction process. Their work is discussed in more detail in Chapter 4.
Once the hand markers have been identified in the video image, it is necessary to convert this raw data into a more readily useful format, such as finger-joint angles. This can be done by matching the visual-position data against a model of the hand and calculating the required values from this model. Again this process is covered in more detail in Chapter 4.
Myron Kreuger (1990) has developed several non-immersive virtual reality systems, which rely on extracting data about the user from a video image.4
Kreuger's systems convert the user's image into a silhouette and then gather data about their motions from this silhouette. Whilst some hand gestures are recognised, this is possible only if they are made away from the body so as to be visible in the silhouette. Therefore this technology is not suitable for gathering sign-language data, in which the hands are often placed on or directly in front of the body.
Research being conducted at Bielefeld University is also attempting to bypass the use of markings on the hands by classifying hand postures directly from a video image by using Local Linear Mapping (LLM) neural networks. So far this research has been based on simulated images generated from a computer model of the mechanics of a human hand. The LLM network is trained on images of hand postures presented at various orientations and learns to classify these postures. When applied to images of real hands the results have been promising although still well short of the accuracy required for this approach to be practical for recognition of real hand gestures (Meyering and Ritter 1992a, Meyering and Ritter 1992b).
4 Non-immersive is a term used by some virtual reality researchers to describe systems
which do not require the user to wear hardware, as opposed to immersive systems which are based on devices such as instrumented gloves and head-mounted displays.