Real-time recognition system of Korean sign language based on elementary components

Sign language is a method of communication for deaf persons. In communication using hand gesture: sign words and manual alphabets are used together. In this paper is proposed a system which recognizes Korean Sign Language(KSL). KSL is composed of Korean sign words and Korean manual alphabets continuously. To recognize meanings of continuous gestures which have no token of beginning and end, this system segments current motion states using speed and change of speed in motions and state automata. To understand the meaning of a gesture, basic component classifiers using Fuzzy Min-Mat Neural Network and fuzzy logic are used. Basic elements of meaning used in this system are 14 hand directions, 23 hand postures, and 14 hand orientations. Meaning of signed gesture is interpreted using basic elements which were recognized by 3 classifiers. This system recognizes 31 Korean manual alphabets and 131 Korean signs in real-time with recognition rate 96.7% for Korean manual alphabets and 94.3% for Korean sign words in case of excluding no recognition case.