List of Figures

• The joints in their hand together with their associated degrees of freedom.
• The tendons that are present in the hand. Again, note the high complexity of the muscles and tendons within the hand, leading to a wide range of possible motions.
• The wide number of possible motions that the hand can make.
• The continuum between Auslan and Signed English. PSE lies in the middle.
• The neutral space for signs. Exact location is not important within this area.
• A picture of the internals of the VPL DataGlove, attached to a Macintosh.
• The Exos Dextrous Hand Master
• The internal structure of a neuron
• A neural network -- a collection of neurons connected together.
• A graphical representation of the decision tree produced by C4.5
• The decision tree discussed so far partition the space as shown.
• An example where the limitations of C4.5 come through. If the concept boundary is non-orthogonal, C4.5 has to make some approximations.
• The eccentricity in the diagram is defined to be MaxDev/StraightPath. The value is positive if it is on our right as we follow the trajectory, and negative if it is on our left.
• Wexelblat's system for use within AHIG.
• The structure of the recurrent neural network used by Murakami and Taguchi.
• The complete Talking Glove system.
• The approach adopted by Kramer to recognise finger spelling.
• The complete GRASP system
• Screen dump of the gloveplay program.
• The Auslan Dictionary definition of same.
• The user starting the sign
• Towards ...
• Meeting fingers ...
• Away ...
• The user indicates the sign is complete.
• The bounding box for the sign same.
• An example of a histogram -- in this case of motion in the x-dimension.
• Graph of the relationship between error and the number of divisions used in making the histograms.
• The possible rotation values coming from the glove and their encoding and classification into histograms.
• Two examples of vectors and the octants they fall into -- one in octant and the other in octant .
• Original data and time-division approximation. Note the clarity of the time-division approximation -- it also manages to reduce the effects of noise.
• The variation in the error rate that occurs with segmentation.
• The effect that the number of samples has on the error rate.
• The effect of the number of samples on the error rate, this time with a logarithmic x-axis.
• The effect of an increasing number of signs on the error rate.
• The effect of an increasing number of signs on the error rate using IBL1, this time with a logarithmic x-axis.
• What can happen when GRASP is trained on one set of users and then tested on another.
• The relationship between the number of samples per sign and the error rate.
• The time per test example and its relationship with the number of signs being learnt.