Gestures and neural networks in human-computer interaction

One approach to improving human-computer communication is to copy aspects of human-human communication. Gesture is a very natural and much-used component of human-human communication but it is very imprecise. The authors present some preliminary work investigating the feasibility of using a neural network to recognize gestures from a small set of structured gestures-letters in sign language. Simulations have been carried out which use a neural network trained to recognize sign language letters. These show a very high level of recognition, even when there is considerable noise in the data. Sign language has been used in these experiments because it is a formalized symbolic gestural language but its interpretation by computer may have some significant applications. At the same time there is also a vast number of other applications which would lend themselves to gestural input on a less formal, but still symbolic basis and some of these are discussed