Dance learning and recognition system based on hidden Markov model. a case study : aceh traditional dance

Abstract

The whole dance of Likok Pulo are modeled by hidden markov model. Dance gestures are cast as hidden discrete states and phrase as a sequence of gestures. For robustness under noisy input of Kinect sensor, an angular representation of the skeleton is designed. A pose of dance is defined by this angular skeleton representation which has been quantified based on range of movement. One unique gesture of dance is defined by sequence of pose and learned and classified by HMM model. The system was implemented using the Matlab and Simulink programming package. Six of dance's gesture classes from the phrase Assalamualaikum has been trained with hundreds of gesture instances recorded by the XBOX Kinect sensor which performed by three of subjects for each gesture class. The classifier system classify the input testing gesture into one of six classes of predefined gesture or one class of undefined gesture. The classifier system has an accuracy of 94.87% for single gesture.