Recognition of signs from a sign language

Consider the task of recognising signs from a sign language using instrumented gloves. The glove provides the information shown in Table 2.2.

Table 2.2: Information provided by gloves.

Channel	Description
x	Hand's position along horizontal axis (parallel with shoulders).
y	Hand's position along vertical axis (parallel with sides of body).
z	Hand's position along lateral axis (towards or away from body).
roll	Hand's orientation along arm axis (i.e. palm facing up/down).
thumb	Thumb bend (0 = straight, 1 = completely bent)
fore	Forefinger bend (0 = straight, 1 = completely bent)
middle	Middle finger bend (0 = straight, 1 = completely bent)
ring	Ring finger bend (0 = straight, 1 = completely bent)

Each instance is labelled with its class, and all of the values are sampled approximately 23 times a second. Each training instance consists of a sequence of measurements. Training instances differ in the number of samples; depending on the duration of the sign itself - some like my are short, while others like computer are longer.

An example training instance is shown in Figure 2.1. This is a recording of the Auslan sign come. The horizontal axis represents the time and the other axes represent normalised values between -1 and 1 for x, y, and z and between 0 and 1 for the remaining values.

**Figure 2.1:** An example of a stream from the sign recognition domain with the label `come`.
$\begin{figure}\begin{center} \leavevmode \epsfxsize =5in \epsfbox{come0.eps}\par\centering\centering\end{center}\end{figure}$

The objective in this domain is: given labelled recordings of different signs, learn to classify an unlabelled instance.