Video-based handshape recognition using a handshape structure model in real time

This paper presents a visual prototype for the recognition of 32 different handshapes used in finger spelling and in German sign language. A signer performs handshapes in front of a single video camera. In order to reach real-time performance the signer wears a coloured glove. The handshapes are recognized by the system with an accuracy of 94%. A description of the overall system is given and an approach to distinguish handshapes by their structure is presented. The lack of information within a recorded image is compensated by the use of a handshape structure model. This model is based on the features of the coloured areas and on relations between these areas.