Fuzzy video-based handshape recognition
Kirsti Grobel, Hermann Hienz · 1996
In this paper a prototype for video-based recognition of 32 handshapes used for fingerspelling and communicating in sign language is presented.Each handshape can be characterized by the constellation of the fingers.Looking at video images the 2D-projection of the constellation is visible, which is described by relations concerning the fingers, one to another and to the ball of the hand.Humans vary in signing and consequently the projection of the handshape changes as well, hence fuzzy logic seems to be a suitable formalism to describe finger relations.A description of the overall system which contains modules for image acquisition, segmentation, handshape representation, feature extraction, and fuzzy classification is given.This paper focuses on the handshape structure model as well as on the classification module that consists of a crisp pre-selection and a fuzzy classifier.To enhance the processing speed the fingers and the ball of the hand are marked with coloured areas on a glove.Only a single video camera is used and a recognition rate better than 94% is reached.