GloveGRASP Gesture Recognition Description GloveGRASP is a C/C++ class library that allows software developers to add highly accurate gesture recognition to their SGI code. GloveGRASP uses advanced pattern matching techniques to ensure highly accurate gesture recognition. Users can set the parameters used in the pattern matching to tune the recognition for their particular gesture set. Recognition rates of more than 95% can be achieved with gesture sets containing as many as a dozen separate gestures. Unlike other gesture recognition systems, GloveGRASP uses context dependent feature based recognition. This means that the same gesture may be interpreted differently for various interaction contexts - so a small gesture set can provide a large range of commands across all the possible interaction contexts. Contexts and the symbols representing each gesture are completely user definable. Allowing context dependent recognition reduces the need to remember dozens of different gestures and produces a very high recognition rate. The GloveGRASP package includes the following features: 5DT 5th Glove SGI device drivers user dependent gesture training one or two handed gesture recognition real-time feature-based continuous and discrete gesture recognition context dependent recognition and dynamic switching of training sets inter-process communication through UNIX TCP/IP sockets GloveGRASP also includes OpenGL and Open Inventor hand models, complete documentation, and C/C++ source code for several sample applications showing how the software libraries can be used, including: a gesture based modeling program gesture navigation through a virtual environment client/sever code for distributed gesture recognition Complete Documentation for GloveGRASP is available in Postscript form or online in HTML. GloveGRASP has been licensed to General Reality and is available as a commercial product.