Gesture-recognition is an important component for many intelligent human–computer interaction applications. For example, a realtime sign-language recognition system would detect and interpret hand gestures. Many vision-based sign-language recognition methods have been proposed over the years with mix results of usability. Some system are limited to recognize only a few gestures, while others require the use of 3D camera to provides depth information to improve recognition accuracy. In this paper, a Kinect-based Taiwanese sign-language recognition system is proposed. Three main features are extracted from the signing gestures, namely hand positions, hand signing direction, and hand shapes. The hand positions are readily available through the input sensor. The signing direction is determined using HMM on trajectory of the hand movement, and a SVM is trained and used to recognize the hand shapes. Experimental results show that the proposed system achieved an 85.14 % recognition rate.
ASJC Scopus subject areas