Instrumented gloves -- gloves equipped with sensors for detecting finger bend, hand position and orientation -- were conceived to allow a more natural interface to computers. However, the extension of their use for recognising sign language, and in this case Auslan (Australian Sign Language), is possible. Several researchers have already explored these possibilities and have successfully achieved finger-spelling recognition with high levels of accuracy, but progress in the recognition of sign language as a whole has been limited.
Most, however, have focussed on the use of expensive hardware and/or complex learning techniques. An attempt is made in this thesis to explore the possibilities using cheap hardware (a PowerGlove worth approximately $100) and computationally simple approaches.
95 discrete Auslan signs were selected, and five different users provided instances of these (there were 6650 samples collected in total). Features were extracted and simple classification such as instance-based learning and symbolic learning techniques were validated. The feature sets were then combined and the overall accuracy tested.
With 95 signs, some selected for their similarity to each other, the accuracy obtained was approximately 80 per cent. It is computationally simple and fast enough to fit on a small portable computer. This compares extremely favourably with the state of the art, with a lexicon almost two and a half times larger than any previous published results.
Potentials for future expansion to a large lexicon system and its expansion to handle inter-signer recognition are explored.
For more information, see the following Web page: