The project involved connecting up a matel power glove, which is a Data Glove produced by matel for games. It was known that certain sequences when sent to the glove could be used to switch it into a higher resolution mode that would be useful for doing real VR experiments with it. Steves project involved finding out how to switch the glove into the high resolution mode, writing a driver for it, a gesture recognition library based on artificial neural networks that would run on out parallel Transputer racks, and some demonstrations of it all working.
1 Disclaimer!I'm finding it very difficult to produce a coherent diagram to show my plans over the next 5 weeks, since many of the jobs that need doing are small, and need not occur in any given order.This text attempts to
describe what I intend to do.
2 Gesture recognitionI intend to have gesture recognition of some kind working within two weeks.My initial attempt at this problem will be to use the quasi-recursive network idea, that is, using one (large) network with a certain amount of feedback to recognise gestures as a function of previous recognised postures and time. If this attempt does not look promising within 4 days, it will be abandoned, and a method using a single large network and a relative-position buffer will be investigated for a maximum of two days. If this seems feasible (I think it might be too slow), I will implement this; if not, I will provide some heuristics to recognise gestures. The system will recognise
the gestures grasp and drop
3 Tool DevelopmentThis section is to take 1 week and will be broken down as follows
4 Demonstration
5 IntegrationVague and fluffy; remaining time integrating with KN's project and Aviary. |